Scheduler: replace setInterval with production-grade cron (BullMQ repeatable jobs)

[parag]

Problem

The routine scheduler uses setInterval(tick, 60_000) — a plain JS timer inside the Node.js process. This works for single-instance dev but has problems in production:

• No persistence — if the server restarts, missed ticks are lost
• No distributed locking — running 2+ instances fires routines twice
• ±60s precision — checks every minute, not at exact cron time
• No catch-up — if the server was down during a scheduled tick, it's silently skipped

Current implementation

packages/@boringos/core/src/scheduler.ts — createRoutineScheduler()
  → setInterval(tick, 60_000)
  → tick() queries active routines, matches cron, fires

Proposed solution

Use BullMQ repeatable jobs when Redis is available:

1. When app.queue(createBullMQQueue({ redis })) is configured, the scheduler should register routines as BullMQ repeatable jobs instead of using setInterval
2. BullMQ handles: cron precision, persistence across restarts, distributed locking (only one instance processes each job), automatic catch-up for missed jobs
3. Fall back to current setInterval when no Redis is configured (dev mode)

Architecture

No Redis (default):
  setInterval(60s) → tick() → match cron → fire  (current behavior, unchanged)

With Redis:
  For each active routine → BullMQ.add(routineId, {}, { repeat: { cron: expression } })
  BullMQ processes → fires routine (workflow or agent wake)
  On routine CRUD → update/remove BullMQ repeatable job

Impact

• packages/@boringos/core/src/scheduler.ts
• packages/@boringos/pipeline (may need repeatable job support)
• packages/@boringos/core/src/boringos.ts (pass queue adapter to scheduler)

Files

• scheduler.ts:16 — createRoutineScheduler()
• scheduler.ts:49 — interval = setInterval(() => tick().catch(() => {}), 60_000)