Get the benchmark tests GitHub action passing again

[mattmccutchen-cci]

The GitHub action for the benchmark tests has been failing for a long time; John said there is a long-standing known problem with one benchmark. We should get the action passing again so its pass/fail status is useful to us.

If one benchmark is failing, Mike suggested we could just remove it. Alternatively, I suggested we could still run it but exclude it from the overall action status (a concept similar to lit's XFAIL) if we find it useful to view the periodic runs on GitHub rather than manually running the benchmark locally as desired.

Tentatively assigning to John.

[mattmccutchen-cci]

John has created a label for issues that block the benchmark tests.

[kyleheadley]

I've just expanded on our filter for bounds errors in a way that will allow us to ignore many more errors, customized with regexes. Simply editing a CSV database will affect results without changing error reporting or removing benchmarks. We should have this discussion again soon, to choose which errors, and which benchmarks, to ignore entirely.