Chat Stream

Stream AI answer via Server-Sent Events (SSE). Same as POST /chat but with real-time token streaming.

SSE 流式发送消息并获取 AI 回答.

Holds no request-scoped DB session. The endpoint takes neither Depends(get_db) nor Depends(get_optional_user) — the latter carries its own get_db yield-dependency, which FastAPI keeps checked out until the StreamingResponse body is fully sent, i.e. the entire 60-120s LLM window, pinning a PG connection per active stream (project_fojin_db_pool_streaming). Instead the raw bearer token is read off the request and the user is resolved INSIDE the generator's short-lived prep session. The generator owns two short-lived sessions (prep + save) and holds none across the LLM I/O.