Tighten-up on Prometheus updates

[anjackson]

As Prometheus updates are done via the event handlers they only get one chance to post metrics: when the task succeeds.

Is it possible to use DEPENDENCY_PRESENT to record when retries show the task is already done?

An alternative would be to only use that for failures, and use a Task superclass to try (and re-try) Prometheus Push Gateway updates as part of the complete check.