Create a benchmark workflow

[celinval]

We don't have any reliable mechanism to assess Kani's performance. This makes trade-off assessment very challenging, since the performance cost of any change is unknown.

Requirements:

• The benchmark suite should include real world examples that are not trivial.
• The report should include at least the following stats:
  • Total runtime
  • Compilation runtime
  • Verification runtime
  • Peak memory consumption (per process)
• Stats should be broken down on a per example base as well as aggregated.
• Workflow allows the comparison between 2 or more commit IDs.
• If the workflow takes too long to run in every PR, implement a mechanism to trigger the workflow on a PR. (E.g.: via performance label).
• The workflow should use the release build.

Next steps:

• [ ] Add more benchmarks under tests/perf.
• [ ] Create a workflow that runs all benchmarks with the release build.

Follow-up work

Once we have a mechanism to measure performance, we should evaluate some things that may either be low hanging fruit or that we suspect can be improved.

[zhassan-aws]

This PR was an initial step in that direction.

[tautschnig]

Should we resolve this now that #2370 has been merged?

[celinval]

@tautschnig I would like to see some improvements in the report before we close this. I was just talking to @karkhaz about this today.