feat: add database-backed scheduler

[winetree94]

Summary

• Add database-backed scheduler mode with durable scheduled-job and background-job storage
• Harden distributed lease acquisition/renewal/completion semantics for multi-instance runners
• Abort cooperative scheduled/background handlers via JobRunContext.signal when lease renewal detects lost ownership
• Improve CAS contention handling so a failed lock attempt continues scanning other due jobs in the same tick
• Add JSON-safe background job payload validation to reject unsupported values before persistence
• Add completed background-job retention cleanup with configurable background_retention_ms (default 7 days)
• Document database scheduler configuration, retention behavior, and instance/lock owner semantics
• Cover scheduler, background queue, standalone config, and lease-loss edge cases with regression tests

Test Plan

• corepack pnpm --filter @tinyrack/tinyauth-server test -- --run src/entrypoints/scheduler/database.test.ts
• corepack pnpm test
• corepack pnpm build
• corepack pnpm biome check .
• git diff --check

Notes

• instanceId remains internally generated by default (hostname:pid:uuid) to avoid lock-owner collisions across replicas; the programmatic override is kept for tests/debugging.
• Lease-loss abort is cooperative: handlers receive an AbortSignal and can stop side effects promptly when renewal returns false; completion is still protected by the existing DB lease fencing.
• Background retention cleanup only deletes terminal jobs (succeeded/failed) with completedAt older than the configured retention window; pending/running jobs are never removed by cleanup.