Resilience Updates

[evantahler]

1. Move the act of poping a job off of a queue to a transaction via watch and multi.

   • This will have speed implications. In a busy system, the occurrences of trying to "pop" while writes are happening is high.
   • We may want to move to a LUA implementation to be truly blocking

2. Add queue.retryStuckJobs() which is a single method to retry jobs which have failed due to the worker timeout.

3. Add options.retryStuckJobs to the Scheduler, to automatically run the above queue.retryStuckJobs() periodically.

If we can gain certainty that a job can't be lost between poping it and working it... and we have ways to retry stuck jobs, we can be rather sure we can't loose any jobs!