ci: Use configurable !benchmark PR comment

[samuelburnham]

Configurable !benchmark PR comment

Uses the reusable workflow from argumentcomputer/ci-workflows#38. This replaces the existing !gpu-benchmark command with
!benchmark, as shown in the following examples:

!benchmark

This will run the default benchmark settings, specified in the workflow file of this PR.
Alternatively:

!benchmark --bench fibonacci --bench end2end --features cuda
LURK_RC=100,600,900
LURK_PERF="fully-sequential"

This will run with each of the input benchmarks, features, and env vars.

Important

• If you want to benchmark with GPU acceleration enabled, you must specify --features cuda in the comment body (see above example). Otherwise, the benchmarks will still run on the same GPU-equipped benchmark machine, but will run using the CPU only.

Details

• The previous !gpu-benchmark syntax will no longer work. The command must start with !benchmark.
• Each benchmark specified with --bench will spawn a new CI job, which means they will each output a separate comment.

The default env vars specified in this PR are

LURK_BENCH_OUTPUT=pr-comment
LURK_PERF=100,600

If identically named env vars are specified in the !benchmark comment, they will overwrite the above defaults.

[huitseeker]

so much work went into this, thanks a lot! 🙏