response-future: complete internal handoffs rejected during shutdown

[dkropachev]

Problem

Several ResponseFuture continuations hand work to Session.submit(). During shutdown, submission can return None or raise after executor teardown, leaving the owning future pending indefinitely when timeout=None.

python-driver/cassandra/cluster.py

Lines 5611 to 5617 in 5651509

	# Known limitation: this retry goes around the RetryPolicy and
	# _query_retries, because both are keyed to the user's statement
	# and this is the driver's own USE. It also has no backoff, so a
	# flapping control connection re-sends USE until the client-side
	# timeout fires instead of failing fast. Routing it through the
	# retry machinery shared with the pooled path would be the fix.
	self.session.submit(self._retry_task, False, host)

python-driver/cassandra/cluster.py

Lines 5914 to 5916 in 5651509

	def _reprepare(self, prepare_message, host, connection, pool):
	cb = partial(self.session.submit, self._execute_after_prepare, host, connection, pool)
	if pool is None and connection is not None and connection.is_control_connection:

The fallback USE, reprepare, schema-refresh, and prepared-query paths share this handoff pattern. This was found while narrowing #392.

Expected behavior

A rejected internal handoff completes its ResponseFuture exactly once with an existing shutdown exception. Internal work must not be counted as a retry-policy decision.

Acceptance criteria

• Shutdown or executor rejection cannot strand these internal continuations.
• Completion remains single-shot when a response races rejection.
• Unit tests cover the affected handoffs with timeout=None.