Autotests: if one of the jobs fail, consider cancelling all other parallel jobs

[rdimaio]

Follow-up from a conversation with Martin. Currently, we run multiple tests in parallel (example). When one of them fails, the others keep running.

To save up on GH workers usage, when one of the parallel tests fail, we could fail all the other tests. However, this might cause annoying problems if there are flaky tests (e.g. #6506)

[voetberg]

Would it also make sense to stop tests on their first fail (adding the -x) option for each runner? This would also free runners faster (not as fast as cancelling the whole suite), and would let the other parallel tests keep running in case of a single flaky test.