Parse, store and compare benchmark results in PRs

[ndr-ds]

Right now we don't run benchmarks on PRs, and we don't really look at the metrics for the ones running on main pushes either, AFAIK. We can probably expand the tool written in #2834 to also parse, store past results somewhere on main pushes, and on PRs parse and compare the benchmark results with the stored ones. And maybe also comment on the PR with the results.

Alternatively you can use this https://github.blog/news-insights/product-news/supercharging-github-actions-with-job-summaries/ to have a summary posted in the job run page, and just have a link to the latest benchmark job run in the PR comment, which will have the summary of the benchmark results.