Under just the right conditions, we could lose a job:

- Job raises an error
- Retry subsystem catches error and tries to create a retry in Redis but this raises a "Redis down" exception
- Processor catches Redis exception and thinks a retry was created
- Redis comes back online just in time for the job to be acknowledged and lost

That's a very specific and rare set of steps but it can happen.

Instead have the Retry subsystem raise a specific error signaling that it created a retry.  There will be three common cases:

1. Job is successful: job is acknowledged.
2. Job fails, retry is created, Processor rescues specific error: job is acknowledged.
3. Sidekiq::Shutdown is raised: job is not acknowledged

Now there is another case:

4. Job fails, retry fails, Processor rescues Exception: job is NOT acknowledged. Sidekiq Pro's super_fetch will rescue the orphaned job at some point in the future.