Add benchmark regression detection

[abrookins]

Summary

Adds benchmark regression detection to fail CI when performance degrades by more than 20%.

Uses benchmark-action/github-action-benchmark to:

• Store benchmark history in the gh-pages branch (under dev/bench)
• Compare PR benchmarks against the baseline from main
• Fail the build if any benchmark regresses by more than 20% (alert-threshold: '120%')
• Comment on PRs when regression is detected
• Only update the baseline when merging to main

This is a follow-up to #798 which added the benchmark test suite.

How it works

1. First merge to main establishes the baseline (stored in gh-pages branch)
2. Subsequent PRs are compared against this baseline
3. If any benchmark is >20% slower, the build fails and a comment is added to the PR
4. When PRs are merged to main, the baseline is updated