Stop the stale rescheduler cloning live and undeliverable matches (#174)

[razam-sherwani]

Closes #174.

Problem

StaleGameMatchRescheduler (@Scheduled(fixedRate=60000)) requeued everything findStaleWaitingMatches returned — and that query selected status IN ('waiting','in_progress') — without cancelling the original. A slow or wedged match got re-enqueued every minute past the 120-min threshold, times_scheduled climbed, and the rescheduler multiplied load under the exact stress it fires in. This was the automatic-load half of last semester's queue, and it recurs for any future tournament.

Fix

• Sweep only waiting. A match becomes in_progress only once a worker emits a "started" update (GameMatchResultHandler:46), i.e. a worker owns it — cloning it is the core defect. A genuinely wedged in_progress match is the per-match watchdog's job (ByteFight-GT/cs3600_2026#35), which this issue is explicitly paired with; that division is intentional, not a gap.
• Bound retries. Once a match has been scheduled gamematch.max-reschedules times (default 5) and is still waiting, mark it failed instead of requeuing, via the existing finalizeMatchResult + failed status. An undeliverable match can no longer loop forever.
• Wire the dead flag. GameMatchProperties.requeueStale was defined and set true in application.yml but read nowhere. The sweep now honours it — a real kill-switch.

No schema change (times_scheduled and MatchStatus.failed already exist).

Tests

New StaleGameMatchReschedulerIT: in_progress-past-threshold is not requeued; waiting-past-threshold is requeued once and times_scheduled bumps; at max-reschedules it flips to failed; requeue-stale: false short-circuits the sweep. GameMatchServiceIT still green.

Verified locally (no CI here): ./gradlew test --tests StaleGameMatchReschedulerIT --tests GameMatchServiceIT on a JDK-17 toolchain, Testcontainers Postgres+RabbitMQ. Independent of #172 and #173.

🤖 Generated with Claude Code