Calls /api/ai/chat, shows the live provider pool, and streams the response from the first healthy backend.
Send a prompt and watch failover headers update in the response.
Output will stream here.