[CON-487] Make state machine error recovery more reliable

[theoilie]

Description

• Fixes onFailure not being recognized due to calling queue.on instead of worker.on (this is most likely what made jobs not re-enqueue properly)
• Makes onError run the same recovery code as onFailure (onError catches "missing lock for job" errors, so we want to recover from those instead of only relying on onFailure)
• Adds global onFailure and onError callbacks using QueueEvents. This should be more reliable than per-worker events because it uses Redis streams instead of pub-sub

Tests

Tested 2 failure scenarios with debugger:

1. Job throws an error in its processor - it recovers and re-enqueues
2. Job throws an error in its onComplete callback - it's not able to recover but logs Fatal: onComplete errored. Cron queues may be broken

Monitoring - How will this change be monitored? Are there sufficient logs / alerts?

• Look out for the following error log: Fatal: onComplete errored. Cron queues may be broken
• Check state machine queues on /health/bull to make sure they all have active or waiting jobs

[SidSethi]

great call on all of these changes