* use peak-ewma instead of head for latency calculation
* Implement some suggested changes from PR
* move latency to new package in workspace root
* fix unit tests which now require peak_latency on Web3Rpc
* Switch to atomics for peak-ewma
This change is to avoid locking from tokio::sync::watch.
* add decay calculation to latency reads in peak-ewma
* Add some tests for peak-ewma
* Sensible latency defaults and not blocking on full
* Cleanup and a couple additional comments
