Uses the workspace's configured LLM integrations. Provide a concrete model to pin the request, or omit it to use auto selection. For streaming, provide request_id, subscribe to llm_chat:{workspace_id}:{request_id} before posting, and set stream=true.