Flaky-test rolling recovery alerts no longer get stuck

Fixed a bug where recovery-enabled rolling flaky-test alerts could silently go dormant on busy projects, stopping them from marking, muting, and recovering flaky tests. The highest-volume alerts, which quarantined the most tests, were the ones that died, while the rest of the automation engine kept running. With this fix, the recovery query scales independently of how many tests an alert has quarantined, an already-wedged alert self-heals on its next evaluation tick, and chronically flaky tests get re-muted while they remain flaky.