feat(multi-site): wire no-inbound relay registration into the real bootstrap flow

sso-manager-node/jump-host already had the relay-automation mechanism
(noInbound/meshIp/publicHost -> theta-proxy route via proxy_client.js,
GET /api/mesh/self on jump-host) but nothing in the actual operator
bring-up flow could ever reach it -- setup.sh, bootstrap/site-join.js,
and setup.env.example had zero wiring for it.

Add bootstrap/site-relay-register.js: reads this spoke's own role from
/config/site.json, logs into the local jump-host as its bootstrap
admin to discover its mesh IP, and registers it with the master. Mesh
peering itself stays a manual step (mint/paste a join token, same
pattern as the site join key), so this runs on every setup.sh
invocation via CFG_SPOKE_NO_INBOUND/CFG_SPOKE_PUBLIC_HOST and is a
no-op ("not meshed yet") until an operator has actually meshed the two
jump-hosts.

Also updates MULTI_SITE_SPEC.md's status table/TODO and the published
mesh.md docs page, which still described this as "designed but not
automated" after the API-level work had already shipped.

docs/MULTI_SITE_SPEC.md

@@ -8,7 +8,7 @@
 > ## Shipped today
 > - **Join, live replication, promotion** (`sso-manager-node`): a spoke joins via a one-time export over a site join key (`POST /api/site/join-keys` / `/export` / `/join`), then registers its own endpoint so the master can push live resync pings on every catalog write — no longer a one-time snapshot. Promotion (`POST /api/directory-admin/site-promote`) coordinates a real handoff, demoting the old master as one action. Identical agent-signing keys ride the same export/resync path. Read [`sso-manager-node/docs/site-join.md`](https://github.com/theta42/theta-directory/blob/master/docs/site-join.md) and `directory_spec.md` §11 for the endpoint-level detail.
 > - **Gateway-to-gateway WireGuard mesh** (`theta-gateway`): real site-to-site tunnels via `POST /api/mesh/register`/`/join`, kernel WireGuard with a userspace `wireguard-go` fallback. Verified with an actual two-container encrypted tunnel passing traffic, not a mock.
-> - **Not yet connected to each other**: the mesh is a transport layer that exists on its own; `sso-manager-node`'s HTTPS-based join/replicate calls don't route over it yet. That wiring, plus the no-inbound relay it would enable (mechanism verified, automation not built — see status table), is the next layer.
+> - **Cross-component routing + no-inbound relay automation**: `sso-manager-node`'s replication traffic now prefers a spoke's mesh IP over the open internet when one is on file (`utils/site_replicate.js`), and a no-inbound spoke's join (`POST /api/site/join` → `/api/site/spokes`) can carry `noInbound`/`meshIp`/`publicHost`, which drives `utils/proxy_client.js` to auto-create the relay route on the master's `theta-proxy` via its existing self-service API token system (reused, not a new credential type). The one piece that stays a manual, out-of-band step is the mesh peering itself (mint a join token on one jump-host, paste it into the other's "Join a mesh" UI) — `theta-suite`'s `bootstrap/site-relay-register.js` (`CFG_SPOKE_NO_INBOUND`/`CFG_SPOKE_PUBLIC_HOST`) picks up from there on the next `setup.sh` run.
 > - **mDNS local-discovery (Linux + Windows)**: shipped and verified — `theta-gateway` announces (`services/mdns_announce.js`), `theta-agent` discovers and applies a hosts-file override, cleanly reverts when the announcement disappears. Linux was verified end-to-end over real multicast; Windows shipped in `theta-agent` v2.2.0 (CRLF-aware hosts override, `ipconfig /flushdns`, and a /32 host-route pin so the WireGuard tunnel can't swallow the direct LAN path). macOS still needs real testing — see the TODO note.
 
 Design scale: a handful of sites (dozen max, 254 hard ceiling — see §4), a few hundred users/hosts total. This is a deliberate, small, trusted-operator deployment, not a hyperscale/adversarial-tenant one — several decisions below (fire-and-forget replication, identical directories) trade blast-radius for simplicity *because* the scale allows it. Don't generalize these choices past that scale without re-deriving them.
@@ -260,19 +260,17 @@ See [`AGENT_LOCAL_DISCOVERY_SPEC.md`](./AGENT_LOCAL_DISCOVERY_SPEC.md) — split
 | Continuous/live replication (vs. one-time export-on-join) | **Shipped** (`sso-manager-node`) — a spoke registers its own endpoint at join time (`POST /api/site/spokes`), and every successful master catalog write fires a fire-and-forget push (`utils/site_replicate.js`) at every registered spoke, which re-pulls a fresh export. Verified end-to-end in `docker-compose.multisite-e2e.yml`. |
 | Identical-directory signing key | **Shipped** — `POST /api/site/export` includes the master's agent-signing key; a spoke adopts it via `agent_keys.adopt()` on join and every resync. OpenBao secret replication *beyond* this one key is still not built. |
 | Coordinated master promotion (demote the old master as one action) | **Shipped** — `POST /api/site/demote` + `site-promote`'s handoff logic. Fixed two real pre-existing bugs while wiring this in: `site-promote`'s god_admin check read a `req.user.groups` field nothing ever populated (permanently 403'd for everyone), and the read-only write-gate 403'd `site-promote` itself before the handler could run. |
-| WireGuard gateway-to-gateway mesh (`theta-gateway`) | **Shipped** — `POST /api/mesh/register`/`/join` (join-token bootstrap), `utils/wg_iface.js` (kernel WireGuard, falls back to userspace `wireguard-go`). Verified with a real two-container test: actual encrypted tunnel, real ICMP traffic across it, 0% loss. This is the mesh transport layer only — nothing in `sso-manager-node`'s replication yet routes traffic *over* it; today's site-to-site HTTPS calls (join/export/resync) still go over whatever network path already reaches the target, same as before this layer existed. |
-| No-inbound-spoke relay (master proxies a spoke with no public IP) | **Mechanism verified, automation not built.** Confirmed with a standalone test (not `theta-proxy`'s actual Lua/Redis engine, which needs its own dedicated pass to wire safely): a spoke with zero published ports, reachable only via its WG mesh IP, served a request that an external client sent to the master's public port — the master terminated the connection and relayed over the tunnel. So the underlying idea works; what's missing is `theta-proxy` automatically creating that relay route when a no-inbound spoke registers (needs a real service-to-service credential between `sso-manager-node` and `theta-proxy`/`theta-gateway` that doesn't exist yet — a new integration, not a small wiring task), and today's HTTPS-based join/replicate still requires the spoke to reach the master's API directly (and vice versa for export), so a spoke with zero inbound *and* zero outbound path still can't join at all. |
+| WireGuard gateway-to-gateway mesh (`theta-gateway`) | **Shipped** — `POST /api/mesh/register`/`/join` (join-token bootstrap), `utils/wg_iface.js` (kernel WireGuard, falls back to userspace `wireguard-go`). Verified with a real two-container test: actual encrypted tunnel, real ICMP traffic across it, 0% loss. `wg_iface.removePeer()` also cleans up the kernel routes `setPeer()` added (verified live: routes present after `setPeer`, gone after `removePeer`, own local route untouched), and `DELETE /api/mesh/gateways/:id` exposes it from the mesh UI. |
+| Cross-component routing (replication over the mesh) | **Shipped** — `utils/site_replicate.js` tries a registered spoke's `meshIp` first (falling back to its public `endpoint` on failure) when pushing resync pings; a spoke with no `meshIp` on file behaves exactly as before. |
+| No-inbound-spoke relay (master proxies a spoke with no public IP) | **Shipped at the API/automation layer, wired into the real bootstrap flow.** `POST /api/site/join`/`/api/site/spokes` accept `noInbound`/`meshIp`/`publicHost` and call `utils/proxy_client.js`, which mints/reuses a `theta-proxy` self-service API token (`prx_...`, OpenBao `secret/integrations/theta-proxy`) and calls the proxy's real Host API to create or update the relay route — verified against a real running `theta-proxy` container (`GET /api/host/:item`'s actual `{item, results: {...}}` response shape, not the flat shape first assumed). `theta-suite`'s `bootstrap/site-relay-register.js` + `CFG_SPOKE_NO_INBOUND`/`CFG_SPOKE_PUBLIC_HOST` (`setup.env.example`) drive it from the operator-facing bring-up flow, re-run automatically on every `setup.sh` invocation until the jump-host mesh IP is discoverable. What's still a manual step, deliberately: the gateway-to-gateway mesh *peering* itself (mint a join token on one jump-host, paste it into the other's UI) — same pattern as minting/pasting a site join key, not something an unattended script should do blind. A spoke with zero inbound *and* zero outbound path still can't join at all (join/export still need the spoke to reach the master's API directly). |
 | mDNS local-discovery (Linux) | **Shipped** — `theta-gateway` announces (`services/mdns_announce.js`, opt-in via `THETA_LOCAL_DISCOVERY_HOSTS`), `theta-agent` discovers and applies a hosts-file override (`local_discovery.go`, opt-in via `prefer_local_directory`). Verified end-to-end with real containers over real multicast: announce → discover → apply → clean revert on disappearance, all confirmed. Caught two real bugs along the way (`mdns.Lookup()`'s IPv6 query aborting the whole lookup even after a valid IPv4 response arrived; `rename()` failing with EBUSY over a bind-mounted `/etc/hosts`, common in every container runtime) — see the commit messages in `theta-agent`. |
 | mDNS local-discovery (Windows) | **Shipped** — `theta-agent` v2.2.0: Windows hosts override (`%SystemRoot%\System32\drivers\etc\hosts`, CRLF-aware, `ipconfig /flushdns` after each change — reachable because the agent runs as a SYSTEM service, so the elevation question resolved in our favor), plus a /32 host-route pin via the owning local interface (`route.exe add ... metric 1`) so the WireGuard mesh tunnel can't swallow the direct LAN path, and a prompt WS reconnect on apply/revert. Tests run the real Windows write path on the Windows CI leg. |
 | mDNS local-discovery (macOS) | Not built — the hosts override compiles on darwin via the shared unix path, but macOS still needs `dscacheutil -flushcache` and real hardware testing (mDNSResponder behavior, hosts-file vs. native Bonjour — see Appendix B §3). Being built on a real macOS VM. |
 
-### TODO — what's actually left, in dependency order
+### TODO — what's actually left
 
-1. **Service-to-service auth model for internal components** (`sso-manager-node` ↔ `theta-proxy`/`theta-gateway`). No unified credential exists today — the site join key (spoke↔master) and the mesh join token (gateway↔gateway) are two separately-invented patterns for the same underlying problem. Items 2 and 3 below both need this; design it once here rather than letting a third integration invent a third pattern.
-2. **Cross-component routing** — route `sso-manager-node`'s HTTPS replication traffic (join/export/resync) over the WireGuard mesh instead of the open internet, now that the mesh exists as its own transport layer. Blocked on #1.
-3. **No-inbound relay automation** — the relay mechanism itself is verified (master terminates + relays to a spoke with zero inbound path — see status table), but nothing in `theta-proxy` creates that route automatically when a no-inbound spoke registers. Blocked on #1.
-4. **Full secret replication** — only the agent-signing key is replicated today. LDAP admin credentials, JWT secrets, and other per-deployment secrets still differ per site, which complicates full disaster recovery.
-5. **Mesh peer cleanup** — `wg_iface.removePeer()` doesn't remove the kernel routes `setPeer()` adds (flagged in code; not yet exercised because nothing removes a mesh peer today).
+1. **Full secret replication** — only the agent-signing key is replicated today. LDAP admin credentials, JWT secrets, and other per-deployment secrets still differ per site, which complicates full disaster recovery. **Paused pending a real-deployment question independent of the code**: this repo's own `conf/secrets.js` was found to contain committed real credentials during this work (LDAP bind, SMTP, VoIP.ms) — see the git-remediation note elsewhere in this repo's history. Building a feature that copies live secrets to additional sites shouldn't proceed until provider-side rotation of those specific credentials is confirmed done; the mechanism itself (generic secret sync, never touching those particular values) can still be designed without that answer.
+2. Service-to-service auth, cross-component routing, no-inbound relay automation, and mesh peer cleanup (the four items formerly listed here) are **done** — see the status table above. What remains genuinely open in that area is documented there inline (mesh peering stays a manual step by design; zero-inbound-and-zero-outbound spokes still can't join).
 
 **mDNS local-discovery, macOS** is deliberately not listed above: the Linux and Windows sides are shipped and verified (`theta-agent` v2.2.0), and macOS is being built on a real macOS VM where the darwin-specific behavior (mDNSResponder/DNS-cache) can actually be tested. Check `theta-agent`'s recent history before assuming it's still open.
 

docs/jump-host/mesh.md

@@ -48,13 +48,17 @@ the hard ceiling this addressing scheme supports.
 - `NET_ADMIN` capability (or equivalent) on the container/host running the
   gateway, to create the WireGuard interface.
 
-## Not yet connected to directory sync
+## Connected to directory sync
 
-This mesh is a networking layer on its own. [Theta Directory's multi-site
-join](../sso/multi-site.html) (catalog + LDAP replication between a master
-and its spokes) does not currently route its traffic over this mesh — the
-two features work independently today. Routing directory sync over the mesh,
-and using the mesh to reach a spoke site with no inbound access of its own,
-are both designed but not yet automated — see the [architecture
+[Theta Directory's multi-site join](../sso/multi-site.html) (catalog + LDAP
+replication between a master and its spokes) prefers this mesh once it's up:
+a spoke that's registered a mesh IP gets its live resync pushes routed over
+the tunnel instead of the open internet, falling back to its public endpoint
+if the mesh path fails. A spoke with no public IP at all can also register as
+no-inbound (`CFG_SPOKE_NO_INBOUND` in `theta-suite`'s `setup.env`) so the
+master auto-creates a relay route through its own `theta-proxy` — the master
+terminates TLS for that spoke's hostname and relays over this mesh. The mesh
+peering itself (this page) stays a manual step on both sides; directory join
+and relay registration pick up from there. See the [architecture
 spec](https://github.com/theta42/theta-suite/blob/master/docs/MULTI_SITE_SPEC.md)
-for current status.
+for the full detail.