Summary:

Changed protocol variance conformance tests to allow errors at either the class definition or the usage site (or both) for cases where the declared variance is wrong. The full discussion is here: #2171

Test plan:

I pinned the pre-existing versions of all the typecheckers because the versioning updates were generating too large of a change, which was muddying the PR. It seemed like a good idea to do the version updates as a separate change, but happy to change follow the latest versions if this is not ok

Results:

• It seems that historical_positional.toml updates are from a recent upstream change (line number shifts)
• mypy, pyright, zuban: Pass the protocol variance (unchanged) and pyrefly still unsupported but the expected results have new line numbers due to our change.