Rolling flaky/reliability alerts now count each test run only once

Tuist’s rolling flaky-test and reliability alerts (FlakyTestsMonitor) now count each test run only once when evaluating thresholds over recent runs. Previously, a run that the flaky detection step re-inserted into the test run table was counted multiple times in the rolling window, which inflated flakiness rates and deflated reliability rates. As a result, on a customer project a burst of activity on 2026-07-02 caused 615 genuinely reliable test cases to be auto-skipped, plus other healthy tests to be muted. With this fix, each run contributes one entry to the window, so skippings and mutings trigger only on real flakiness or reliability drops, matching what the dashboard shows. The fix applies read-time, so no backfill was required.