Fixes hangs when Python future notifier is enabled

[pentschev]

The test_many_servers_many_clients has been occasionally hanging for a long time when Python future notifier is enabled. Since Python 3.13.12, it started hanging almost 100% of the runs. Upon some investigation, it seems the hang comes from the notifier thread in notifier_thread.py:

1. With Python future enabled, completed UCX operations are finished by the notifier: the notifier thread schedules _run_request_notifier(worker) on the event loop via run_coroutine_threadsafe. That coroutine calls worker.run_request_notifier(), which sets the asyncio futures that ep.send()/ep.recv() are awaiting.
2. The notifier thread then does task.result(0.01) and, on timeout, calls task.cancel(). So whenever the coroutine doesn’t finish within 10 ms, it is cancelled.
3. Under Python 3.13.12, asyncio’s callback ordering changed (asyncio.run_coroutine_threadsafe leaves underlying cancelled asyncio task running python/cpython#105836 / gh-105836: Fix asyncio.run_coroutine_threadsafe leaving underlying cancelled asyncio task running python/cpython#141696). The event loop often runs that scheduled coroutine later, so the 10 ms timeout is hit more often. The notifier thread then cancels the task, run_request_notifier() never runs, the asyncio futures for completed UCX requests are never set, and the test hangs in send/recv or in wait_listener_client_handlers.

This resolves the persistent hang in Python 3.13.12, and may as well resolve the hangs that occurred occasionally in those tests as well.

Closes #586

[jameslamb]

Great investigation and very clear description, thanks!

[jameslamb]

I'll admin-merge this for you since only check-nightly-ci is failing. I'll work on getting that fixed today: rapidsai/shared-actions#94

[pentschev]

Thanks @wence- for review, and @jameslamb for review and also admin-merging, appreciate it!