In the past months pretty much all test runs have lots of VM boot failures, retries, and spurious failures. This is most likely because our OpenStack instances have more noisy neighbors, and our VMs get descheduled too often. This is totally unpredictable and really hard to compensate, but we have to do something about it.

Make run-tests scheduler more adaptive and resilient against these failures, so that a fedora-43/devel run has a high chance of passing and only a few automatic retries.