Cross-run flaky detection now flags only the failing run as flaky

Tuist Server’s cross-run flaky detection now marks only the failing run as flaky when a test passes and fails on the same commit, so the passing runs in that group are no longer counted as flaky. This fixes inflated flaky_run_count and flakiness_rate metrics, which could cause auto-quarantine rules to trigger too aggressively and could quarantine entire suites after a retry or flake-check burst. The flaky group views still show the full pass/fail breakdown for the commit by fetching the complete run history, so the Flaky Runs tab and run detail pages remain accurate. Historical data is not backfilled, so existing inflated counts age out naturally within their evaluation window. Repetition-based flakiness within a single run is unchanged.