Shut down scheduler on unrecoverable exceptions

[hendrikmakait]

Several deadlocks (e.g., #8702, #8787) have been caused by exceptions during transition logic, stimuli-handling, or stealing. Instead of deadlocking and keeping a cluster alive indefinitely, we should force-close the scheduler instead similar to what we do on the worker (#6206).