add c++ wrapper for prometheus & common metrics

[jkarneges]

Currently, StatsManager will construct an internally-owned PrometheusServer when prometheus metrics are enabled. This works as long as there is only one StatsManager instance in the whole process. However, the proxy component can be configured to have multiple workers, in which case multiple StatsManager instances will be created, one per worker. These configuration options are in conflict. It is not possible to enable multiple workers and prometheus metrics at the same time, since only one worker would be able to bind the configured port.

To make prometheus metrics work when there are multiple StatsManager instances, we need to rethink this. Either each instance needs to bind its own port, or multiple instances need to be able to share the same port. I think the latter approach is tidier, as it allows the worker count to be tuned without impacting scraper configurations.

Toward that end, this PR adds C++ wrappers around the prometheus and common metrics FFIs, to make it easy to construct these things outside of StatsManager. They are not used anywhere yet, but the plan is for applications to create one PrometheusServer and one CommonMetrics instance, and then share the CommonMetrics instance with one or more StatsManager instances.