Change default initialization of PrometheusHandle

[Ptrskay3]

Addresses #51, based on metrics-rs/metrics#467.

This change prevents unbounded memory growth of metrics histograms by swapping PrometheusHandle::install_recorder to PrometheusHandle::build, that allows the upkeep task to be created, and the data is properly drained.

Other changes:

• Bump metrics to 0.23, metrics-exporter-prometheus to 0.15.
• Document current MSRV

TODO:

• Figure out what to do when push-gateway is enabled.

[Ptrskay3]

Finally found the time to fix this issue. I tested the changes locally, and it no longer produces unbounded memory growth for me. Nonetheless, it'd be great to hear some feedback from someone else too before merging. Also, does it make sense to specify a default 5 seconds upkeep_timeout for most users?

cc @gauravkumar37, @tobz, maybe @yanns

[yanns]

I really appreciate this change, thanks!
In my company, for other services in different languages, we are using a different approach. Whenever the metrics are fetched, we use that information to reset the metrics.
By using a timeout, I guess that one should use a time that is more than the scrape interval (1 minute in our case)
I can check this change later to see how it behaves.

[tobz]

To my eyes, this looks right. 👍🏻

[Ptrskay3]

I'll go ahead and merge this as-is. I'll get back to handling the push gateway feature in a different PR.