Refactor: puma feature category metrics

Before this, we were recording feature category metrics for the puma component based on the promSourceSLIs aggregation set. This means that we'd have a 1m burn rate, which the feature categories don't need.

Because of this adding burnRates to the aggregation set rates would not generate the rules for the puma component.

This refactors this to use the featureCategorySourceSLIs as the source of truth for the generation of this aggregation. Similar to how we generate the feature category recordings for Sidekiq.

This also means that we don't need to have these recordings inside the key metric groups. Keeping that group smaller.

Because of that, we also have less rules to evaluate: instead of recording per service and selecting based on the type label, we record them and include the type as an aggregation label.

A precursor to !4135 (closed)

These recordings hopefully going to be replaced by the new rails_requests component, but since stage groups are opting in, this won't happen tomorrow.