Fix retry logic to prevent setting exception if future is already done

[ofekby]

This pull request makes a small but important change to the retry logic in the __retry method of the src/aioice/stun.py file. The change ensures that the TransactionTimeout exception is not redundantly set if the future is already completed.

• src/aioice/stun.py: Modified the condition in the __retry method to check if the future (self.__future) is not already done before setting the TransactionTimeout exception. This prevents unnecessary operations and potential errors.

We experience this case in production here is the stacktrace:

Traceback (most recent call last):
  File "uvloop/cbhandles.pyx", line 253, in uvloop.loop.TimerHandle._run
  File "[masked]/aioice/stun.py", line 330, in __retry
    self.__future.set_exception(TransactionTimeout())
asyncio.exceptions.InvalidStateError: invalid state

[jlaine]

This looks like a good catch, could you please prepare a unit test which exercises this scenario?

[ofekby]

This looks like a good catch, could you please prepare a unit test which exercises this scenario?

Thanks! @jlaine
Added a test to check the race condition where the message is resolved right before the retry.
Fortunately, the event loop behaves deterministically here - __retry runs before the cancellation, so the bug is reliably reproduced.

The test fails consistently when the future-guard fix is commented out, confirming that the test reliably captures the issue.

To detect the raised exception (InvalidStateError, previously unhandled), I added CollectExceptionsHandler, new_collect_exceptions_handler, and detect_exceptions_in_loop to keep the test clean.