butler metrics needs to be like guy pearce in memento

there is an issue with butler where if you have to manager sections (eg: prometheus, alertmanager), and butler will create prometheus metrics for both of those.

If you remove the configuration for one of the managers, then the prometheus metrics will still hang about, and that's bad.

@kirharri suggests that at each config retrieval iteration, then blow away all the metrics, and re-write them.

let's sort it out

Start by tracing Butler's configuration retrieval iteration and the code that creates Prometheus metrics for each manager section. Done means metrics for removed manager configurations no longer remain, while metrics for active sections continue to be exposed.

Written by the indexing model from the issue text.