Make "expeditive" pipelines way faster

Given the following, I think pipeline:expedite pipelines should be way faster:

• only master:broken MR are allowed to have pipeline:expedite, which means master is already broken and the priority is to fix it
• in most cases, pipeline:expedite MRs are reverts of MRs that were merged a few hours earlier, so the risk of breaking master even more with the revert is minimal compared to keeping master broken while pipeline:expedite pipeline is running
• the merge commit will trigger a complete pipeline anyway afterward, so in the worst case, master will stay broken after merge 🤷🏼

What I propose is to reduce the number of jobs we run in pipeline:expedite pipelines:

• only run predictive jobs (don't run full rspec after approval nor foss-impact)
• don't run permutation jobs (as-if-foss, single-db, es8 etc.)
• don't run db:* jobs
• don't run jest jobs
• we could even go further and only run linting and gitlab:setup

Example of a current pipeline:expedite MR where we had 606 jobs: https://gitlab.com/gitlab-org/gitlab/-/pipelines/834222144 (finished in 84 minutes).

Example of the new pipeline:expedite behavior I propose, with 27 jobs: https://gitlab.com/gitlab-org/gitlab/-/pipelines/837674691 (finished in 9 minutes). 🚀