ReplicationCoordinatorExternalStateImpl::shutdown() must not hold _threadMutex while waiting for _taskExecutor

ReplicationCoordinatorExternalStateImpl::shutdown() holds _threadMutex while waiting for _taskExecutor. This is problematic, since the _taskExecutor's function takes database locks, so we can wait for database locks while holding _threadMutex. This can lead to deadlock, since we can wait for _threadMutex while holding ReplicationCoordinatorImpl::_mutex, and we can wait for ReplicationCoordinatorImpl::_mutex while holding database locks.