Behaviour wrapping the req_llm streaming call.
Exists so that Planck.AI.Adapter can be tested without making real HTTP
requests. In production, Planck.AI.ReqLLM delegates directly to
ReqLLM.stream_text/3. In tests, Planck.AI.MockReqLLM is injected via
application config.