Compatibility tests should not block PRs

[Gudahtt]

The "compatibility tests" are very helpful for discovering unexpected breaking changes in our dependencies. However, they rarely are caused by changes in individual PRs, but rather indicate breakages in dependencies. As such, they aren't useful as a PR status check.

Ideally PR status checks would be solely focused on catching regressions caused by the PR itself, to avoid blocking unrelated work and to avoid sending contributors on wild goose chases.

We could accomplish this by only running the compatibility tests on main. Moving them to a separate workflow would also make the results easier to understand for the team (e.g. so we don't think that our main regression suite is broken).

[Gudahtt]

See here for an example of how these tests can be useful, but also can block unrelated work: https://consensys.slack.com/archives/C1L7H42BT/p1755589196749669