Performance regressions are not caught by CI

[gakonst]

High level idea:

We should consider adding some basic profiling to geth in the CI and upload a benchmark artifact that can we compare over time. If some benchmark gets worse than we can have an optional check fail.

I think we could run geth with prof enabled ( --pprof --cpuprofile cpu.prof), and have a script that generates the graph that @tynes generated in #492 run at the end of the integration tests?

Originally posted by @snario in https://github.com/ethereum-optimism/optimism/issues/492#issuecomment-822815011