llama-mmap : avoid a second full-size copy of each tensor with direct-io (#29749)

Assisted-by: Claude

Co-authored-by: Pranesh Gonegandla <pgonegandla@nvidia.com>

src/llama-model-loader.cpp

@@ -1516,8 +1516,7 @@ bool llama_model_loader::load_all_data(
     }
 
     // Buffer size: balance between memory usage and I/O efficiency
-    // 64MB works well for NVMe drives
-    const size_t buffer_size = alignment != 1 ? 64 * 1024 * 1024 + 2 * alignment : 1 * 1024 * 1024;
+    const size_t buffer_size = alignment != 1 ? LLAMA_DIRECT_IO_BUFFER_SIZE + 2 * alignment : 1 * 1024 * 1024;
 
     std::vector<ggml_backend_buffer_t> host_buffers;
     std::vector<ggml_backend_event_t> events;