Allow jobs to fail and not be retried

[shawnrushefsky]

This behavior would be essentially for input validation type errors where you can be 100% sure the job cannot succeed on a different machine. This behavior would also not count towards the self-destruct counters the worker keeps.