Benchmark CI failure/improvement

[sbryngelson]

Right now the benchmark CI workflow prints useful information about grind time and runtime to the CI output. This CI workflow always passes, so long as the code executes.

I would like the CI to fail if the master and PR runtime or grind times are sufficiently different (within some tolerance). Without this, it is easy to have a performance regression but not see it because one did not check CI carefully enough.