feat(phase2): complete openai api surfaces 2.3-2.7

- 2.3: unknown model errors (400 + RFC 9457 problem+json with valid_models)
- 2.5: GET /v1/models endpoint (derived from config, not hardcoded)
- 2.6: POST /v1/embeddings passthrough (body-based dispatch, no rewrite)
- 2.7: POST /v1/rerank with path rewrite (/v1/rerank → /rerank)
- wire proxy.Handler in main.go (was using dummy handler)
- 140+ tests passing, race detector clean
- all requests: client → nginx → gateway → upstreams
- ready for config deployment to go live

testdata/config/harness-with-models.yaml

@@ -0,0 +1,26 @@
+# Local development harness config with model registry for body-based dispatch.
+# This config is used for testing task 2.2 (body-based dispatch) and task 2.3 (error handling).
+# Every upstream points at the single stub server on loopback 127.0.0.1:9080.
+
+routes: []
+
+models:
+  - name: "reasoning"
+    address: "127.0.0.1:9080"
+    path: "/v1/chat/completions"
+  
+  - name: "ornith:35b"
+    address: "127.0.0.1:9080"
+    path: "/v1/chat/completions"
+  
+  - name: "qwen2.5:3b-instruct"
+    address: "127.0.0.1:9080"
+    path: "/v1/chat/completions"
+  
+  - name: "nomic-ai/nomic-embed-text-v2-moe"
+    address: "127.0.0.1:9080"
+    path: "/v1/embeddings"
+  
+  - name: "BAAI/bge-reranker-base"
+    address: "127.0.0.1:9080"
+    path: "/v1/rerank"