Add a working comparative GPU benchmark workflow/action

There is enough common ground between https://github.com/lurk-lab/lurk-rs/blob/main/.github/workflows/merge-tests.yml#L69 and https://github.com/lurk-lab/arecibo/pull/238 that we should be able to factor out some if not all steps into a reusable workflow or action.

This will help us extend GPU benchmarks to other performance-centric repos in the future like https://github.com/lurk-lab/grumpkin-msm

Compare the steps around line 69 of .github/workflows/merge-tests.yml with the workflow discussed in lurk-lab/arecibo PR 238. Identify the common GPU benchmark pieces and determine how they can be exposed as a reusable workflow or action; done means another performance-focused repository such as grumpkin-msm can use it for comparative GPU benchmarks.

Written by the indexing model from the issue text.