Wire Argu.Benchmarks into CI

[dimension-zero]

Follow-up to #313.

The benchmark harness added in #313 is most useful when results are surfaced regularly rather than living as a project that contributors have to remember to run by hand. Wiring it into CI would:

• Catch perf regressions on PRs against master.
• Make the non-zero parsing cost visible over time, in line with @bartelink's point in feat(benchmarks): Add BenchmarkDotNet harness #313.
• Encourage iterative perf work by displaying numbers alongside the build.

Open questions for whoever picks this up:

• Trigger: every PR, only PRs touching src/Argu, or scheduled nightly on master?
• Reporting: post a comment with the BenchmarkDotNet summary, attach the artifacts, or publish to GitHub Pages / external dashboard?
• Comparison: baseline against master HEAD on PR runs, or just publish absolute numbers?
• Runner: standard GitHub-hosted runners are noisy; do we accept that, or pin a self-hosted runner for stability?

A reasonable first cut: GitHub-hosted runner, run on PRs that touch src/Argu/** or benchmarks/**, post the summary table as a sticky PR comment.

[bartelink]

I'd personally be happy with just per-PR absolute numbers and/or anything you imply by the 'reasonable first cut'. I'll merge 313 in the next few days with ot without that added to it, up to you...

[dimension-zero]

Totally happy with whatever you decide ;o)