Review Zync Prometheus metrics on Que jobs

Ops have requested us to review some of the new Prometheus metrics of Zync, so que jobs are not accounted twice in different types of que_jobs_scheduled_total

See: https://github.com/3scale/platform/issues/230

Summary of the improvements requested:

1. Make "scheduled" not to include failed jobs scheduled for retry
2. Make "failed" to count failed jobs scheduled for retry and not expired ones
3. Introduce a new type "expired" for failed jobs that already ran out of attempts to retry and therefore won't be retried again
4. Remove the type "retried" since it seems not be updated by Que, which handles retries based on the value of the error_count column only.