fix(pg): alert_history retention never ran; resolve state lost on restart

[cursor]

Bug and impact

1. Retention never deleted alert_history rows — cleanupAlertHistory issued DELETE … WHERE ts / created_at, but bootstrap only creates fired_at. Errors were discarded, so Retention.AlertHistoryDays had no effect and the table grew without bound on PG deployments.
2. Resolved alerts reappear as firing after restart — resolveAlertRecord updated only the resolved_at column; loadRecentAlerts rebuilt memory from JSONB (status=firing, resolved_at=0). /api/v1/alerts/history?status=firing showed ghosts; MTTR/history reporting was wrong. Resolve also matched PG BIGSERIAL id against the in-memory alertSeq, which diverges after a failed insert.

Root cause

Stale cleanup SQL against the wrong columns; resolve/load asymmetry between the relational column and JSONB payload; async insert + id-based update.

Fix

• Delete by fired_at < cut and log failures.
• Resolve by key and update both resolved_at and JSONB status/resolved_at.
• Load overlays the relational resolved_at onto JSONB records.
• Persist alert history insert/resolve synchronously after releasing the store lock so resolve cannot race a missing row.
• Drop orphan test refs (licenseResetForTest, classicAppJS) so ./cmd/server tests compile after license/Vue v2 removal.

Validation

• go test -count=1 -run 'TestCleanupAlertHistoryUsesFiredAt|TestResolveAlertRecord|TestLoadRecentAlerts|TestStoreAlertHistory|TestMetricsEndpointAuthAndPayload|TestSweepSkipsInflight' ./cmd/server/
• go build ./cmd/server ./cmd/agent