We had similar issues with integration tests, and made them a separate jenkins job that didn't trigger automatically, but gitlab was still configured to require them to pass for merge. We would kick it off manually only after all other code review was complete. Then the only cases where we had to re-run it were the same cases where it would have failed in master if we only ran the test there, but it saved us the hassle of reverting or feeling pressured to get hotfixes into master quickly.