The bundle size diff message is using the wrong point of comparision,
leading to misleading results on feature branches that have been
merged with `develop` since they were created.
When this feature was introduced, we went back and forth a few times on
what we should be comparing the branch with to get an accurate bundle
size comparison.
The first attempt used `develop` as the point of comparison, but that
didn't work because it was a moving target, and because it didn't
reflect the changes made on this branch. As bundle increases or
decreases were added on `develop`, they would alter the diff on each
feature PR.
Then we chose to use the fork-point of the branch, the commit of
`develop` that the branch forked off of. This works for feature
branches that don't merge in `develop`. But the minute `develop` gets
merged in, then unrelated changes on `develop` affect the measurement.
The _most recent_ commit from `develop` on the current branch is a
better comparison. Any difference between this commit and the feature
branch in terms of bundle size would be attributable to the feature
branch changes. This is what `merge-base` gives us.