bench bot subweight output tuning

[bkontur]

Problem

After running the bench bot, we execute subweight to compare the fresh weight files with the base branch. Currently, this base branch is hardcoded as refs/remotes/origin/master for polkadot-sdk and refs/remotes/origin/main for polkadot-fellows.

This setup works fine when the bench bot is used on a pull request targeting master or main within the same repository. However, it has several downsides, as it can produce misleading results:

• If the PR originates from a fork, the comparison is made against the fork’s master or main, which is incorrect.
• If the PR’s target/base branch is different from master or main, issues arise. For example, if I have one PR (origin/pr1 → origin/master) and then another PR (origin/pr2 → origin/pr1), running the bench bot in the second PR should compare origin/pr2 with origin/pr1.

Potential Solution

Instead of using hardcoded references to refs/remotes/origin/master (polkadot-sdk) and refs/remotes/origin/main (polkadot-fellows), we should dynamically determine the PR’s actual target/base branch.

This can potentially be achieved using some git remote add scripting along with fetching PR data via the GitHub REST API, as referenced here.

[bkontur]

cc: @mordamax