Benchmark pipeline improvements

[Kamirus]

The current approach is not prepared for scaling the number of benchmarks. As we add more and more benchmarks the output becomes too long and unreadable and the CI time increases:

• Avoid running benchmarks for PRs that have unrelated changes
• Skip showing benchmarks that do not differ compared to main -- show only changed and newly added
• Evaluate more sophisticated workflows instead of using the markdown table output from mops bench, e.g. github-action-benchmark

Ideal scenario: avoid relying on custom scripts to lower the maintenance burden.