context

motivation: we want a running top-of-the-stack measure of how long it takes to send messages so that we can measure whether perf improvements have an impact on decreasing message lag
notes on things i changed since spec'ed:
- i renamed this metric to "metric roundtrip" since it more accurately names what we are measuring (to measure "lag" we'd want to subtract healthy roundtrip time from observed. that's different!)
- after inspecting docs on gauge, i realized it's sort of flawed for what we're trying to do. left notes on why i think that, and added a histogram

changes

record whenSent in state when registering callback, subtract whenSent from current time in callback handler to derrive round trip time
for now, record metric in both histogram and (likely flawed) gauge so we can see what yields most useful data (but note why gauge approach seems flawed)

restore prometheus/grafana containers to dev docker compose (note: it would be nice to have a "quiet" option that excludes them!)
leave localdev network available in comments for running tests behind VPN (pending a solution that would work both for dev and gitlab CI runner...)