RuntimeError: Task attached to a different loop

Root cause

A FastAPI startup fixture initialized app.state.dispatcher and called asyncio.create_task() while pytest-asyncio was still using a session-scoped loop. Later tests ran under a reset function loop, so the dispatcher's queue and worker future belonged to the old loop and enqueue() awaited them from the new loop.

Minimal fix

Make the async app/client fixture function-scoped, create the dispatcher only inside the active test event loop, and explicitly await FastAPI lifespan startup before yielding the httpx AsyncClient. Ensure shutdown cancels and awaits the background worker before pytest-asyncio resets the loop.