Do not run benchcomp in every PR

[celinval]

Description of changes:

The performance benchmark PR is taking a lot of our CI time and many times it's not necessary. For example, benchmark job over 2h while all the other runs ran in less than 40min in this run

This PR moves the performance benchmark to a separate workflow that only runs on pushes to main / release tags and PR's that are labeled with 'Z-BenchCI'

Resolved issues:

Related RFC:

Call-outs:

We still need to fix the release tag runs to compare against previous release. We haven't had this working before, so I hope it's not a blocker for this PR. I could just remove the push to tag trigger for now if people prefer or create an issue to finish supporting release comparison.

I created a label for now, but we can rename it or remove it if this PR doesn't get accepted.

Testing:

• How is this change tested? I tested this in my fork. Here is the job run for a PR where I applied the label and here is the run after I pushed to my fork main branch.

• Is this a refactor change? It's a CI refactoring change

Checklist

• Each commit message has a non-empty body, explaining why the change was made
• Methods or procedures are documented
• Regression or unit tests are included, or existing tests cover the modified code
• My PR is restricted to a single feature or bugfix

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 and MIT licenses.

[jaisnan]

Thanks! I agree with the change. Should we document the need for the label in our dev-documentation guide?