Drop the DR suite's CancelledError workaround once aiokafka fixes Fetcher.close() #116

Blocked upstream by aio-libs/aiokafka#1210. Nothing to do until a released aiokafka contains a fix.

Background

aiokafka 0.14.0's AIOKafkaConsumer.stop() raises asyncio.CancelledError when it is called while no broker is reachable. Fetcher.close() cancels each pending fetch task and then awaits it without suppressing the cancellation, and a fetch task that is sleeping in its retry backoff ends with CancelledError. The upstream issue has the details and a standalone repro.

This library's code is not affected: it never calls broker.stop(). The DR suite added in #114 does. Its shutdown-during-outage scenario in tests_dr/test_dr.py catches CancelledError around broker.stop() so the scenario can go on to check redelivery after the cluster comes back.

What to do once aiokafka ships a fix

• Confirm uv.lock resolves to an aiokafka release that contains the fix.
• In tests_dr/test_dr.py, catch Exception only around broker.stop() in the shutdown-during-outage scenario, and delete the comment that points at the aiokafka bug.
• Run just test-dr and check that the scenario passes with broker_stop_error reported as None.