This will shorten the feedback loop for MRs while they're in development. But we still run all the tests on `main` and for merge train pipelines.
