feat: 맥미니 MLX 연동 — OpenAI-compat 프록시 + 모델 배치 정정

- proxy_openai.py 추가: MLX 서버 SSE 패스스루
- chat.py: openai-compat 백엔드 타입 라우팅 추가
- backends.json: GPU=embed(bge-m3)만, 맥미니MLX=채팅(qwen3.5:35b-a3b)
- LAN IP(192.168.1.122) 사용 (같은 서브넷, Tailscale 불필요)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

hub-api/routers/chat.py

@@ -5,7 +5,7 @@ from fastapi.responses import JSONResponse, StreamingResponse
 from pydantic import BaseModel
 
 from middleware.rate_limit import check_backend_rate_limit
-from services import proxy_ollama
+from services import proxy_ollama, proxy_openai
 from services.registry import registry
 
 router = APIRouter(prefix="/v1", tags=["chat"])
@@ -79,7 +79,24 @@ async def chat_completions(body: ChatRequest, request: Request):
             )
             return JSONResponse(content=result)
 
-    # Placeholder for other backend types
+    if backend.type == "openai-compat":
+        if body.stream:
+            return StreamingResponse(
+                proxy_openai.stream_chat(
+                    backend.url, body.model, messages, **kwargs
+                ),
+                media_type="text/event-stream",
+                headers={
+                    "Cache-Control": "no-cache",
+                    "X-Accel-Buffering": "no",
+                },
+            )
+        else:
+            result = await proxy_openai.complete_chat(
+                backend.url, body.model, messages, **kwargs
+            )
+            return JSONResponse(content=result)
+
     raise HTTPException(
         status_code=501,
         detail={