RFC: collect stackdriver metrics globally and include them with the process metrics

[hsyed-dojo]

Context in #173.

@SuperQ we are quite close to a single deployment design where we pass filter params to the scrape endpoint.

However, when using this approach the innerHandler registers process scoped metrics (e.g: stackdriver_monitoring_api_calls_total) with the throwaway registry. These are not globally cumulative and they get reset with each scrape.

We would need these to be globally cumulative and (preferably only) available with the process metrics.

What do you think of these changes ?

I have time to work on this this week.

RFC:

• Should the stackdriver scrape endpoint still include these metrics ? If not this is a breaking change.

[antoinedeschenes]

I agree with moving stackdriver_monitoring_ metrics to the process path, right now we deduplicate stackdriver metrics overlapping across multiple exporter instances, but these metrics make no sense being instance-specific