scheduler: complete queued work when executor rejects submission

[dkropachev]

Problem

After #392, normal explicit and interpreter shutdown stop the scheduler before executor teardown. An externally shut-down or broken executor can still reject _Scheduler.run() after the task has been removed from the queue.

python-driver/cassandra/cluster.py

Lines 5118 to 5135 in 5651509

	def run(self):
	while True:
	if self.is_shutdown:
	return
	try:
	while True:
	run_at, i, task = self._queue.get(block=True, timeout=None)
	if self.is_shutdown:
	if task:
	log.debug("Not executing scheduled task due to Scheduler shutdown")
	return
	if run_at <= time.time():
	self._scheduled_tasks.discard(task)
	fn, args, kwargs = task
	kwargs = dict(kwargs)
	future = self._executor.submit(fn, *args, **kwargs)
	future.add_done_callback(self._log_if_failed)

The scheduler thread then exits without completing the claimed task or draining shutdown callbacks, which can leave a waiting future pending. Runtimes without the early threading hook have the same fallback gap.

Expected behavior

Executor rejection makes the scheduler terminal and completes callbacks for the claimed and queued tasks exactly once.

Acceptance criteria

• The claimed task is not lost when submit() rejects it.
• Queued shutdown callbacks are drained once.
• Concurrent and reentrant shutdown remain deadlock-free.
• Unexpected executor failure is logged.