Fix 404: map 'code-memory' model name to real Ollama model

Open WebUI sends the proxy model name to our endpoint; passing it
directly to Ollama caused 404. Now stripped before forwarding.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

main.py

@@ -127,13 +127,15 @@ async def chat_completions(req: ChatRequest, bg: BackgroundTasks):
     messages = await build_context(req)
     user_content = next((m.content for m in reversed(req.messages)
                          if m.role == "user"), "")
+    # use real Ollama model, not our proxy name
+    ollama_model = None if req.model == "code-memory" else req.model
 
     if req.stream:
         collected: list[str] = []
 
         async def gen() -> AsyncIterator[bytes]:
             try:
-                async for chunk in oll.chat_stream(messages, model=req.model):
+                async for chunk in oll.chat_stream(messages, model=ollama_model):
                     collected.append(chunk)
                     data = json.dumps({"choices": [{"delta": {"content": chunk},
                                                    "finish_reason": None}]})
@@ -153,7 +155,7 @@ async def chat_completions(req: ChatRequest, bg: BackgroundTasks):
         bg.add_task(_save_stream)
         return StreamingResponse(gen(), media_type="text/event-stream")
 
-    response_text = await oll.chat(messages, model=req.model)
+    response_text = await oll.chat(messages, model=ollama_model)
 
     # background: save turns + extract facts
     async def _save():