refactor(worker): move coexist jobs to dedicated heavy worker

[viethung0o0]

Summary

• Moves the 5 coexist historical-sync jobs (coexistWhatsappBuffer/Flush, coexistMessengerSync, coexistInstagramSync, coexistAttachmentDownload) off the integration queue onto a new workload-class heavy queue with its own worker, so long-running Graph-API sync chunks no longer occupy the concurrency slots that real-time chat jobs (incomingMessage, sendFlow) need.
• The heavy queue is a generic tier for long-lock, throughput-oriented jobs — coexist is its first tenant; future heavy workloads (e.g. import backfills) join it instead of spawning new workers.
• The integration worker keeps a small forward-only shim for legacy coexist jobs still in bull:integration after deploy (forwards them to the heavy queue preserving jobId and options); scripts/check-coexist-drain.mts gates a follow-up PR that deletes the shim once the old queue is clean. Its stalledInterval override is also removed so crashed-worker jobs are re-picked in seconds.

Changes

• packages/worker-config/src/queues/heavy/ — new heavy queue module: HeavyJobAction, HeavyJobData (+ runtime heavyJobDataSchema), heavyQueue; the 5 actions removed from the integration queue module.
• apps/worker/src/heavy/ — new worker (blocked-owner guard, audit context, 10-min lock sized for sync chunks, HEAVY_WORKER_CONCURRENCY env) with the coexist handlers moved under handlers/coexist/ (git renames, history preserved); all coexist follow-up enqueues retarget heavyQueue (updateContactAvatar fan-out stays on integration).
• apps/worker/src/integration/worker.ts — coexist cases/handler imports removed; forward-only legacy shim added; lock comment re-derived from CHAT_JOB_WAIT_TIMEOUT_MS; stalledInterval override dropped.
• apps/worker/src/schedule/handlers/scan-coexist-runs.ts, packages/business/src/coexist/service.ts — enqueue/strategy types retarget the heavy queue (jobId schemes unchanged).
• packages/sdk + integrations/whatsapp — HandleRequestProps gains a generic optional heavyQueue; the WhatsApp webhook enqueues the coexist buffer through it and logs an error if it is missing; both builder webhook routes inject it.
• apps/builder/src/app/developer/queues/** — heavy queue registered in the queues dashboard.
• scripts/check-coexist-drain.mts — read-only drain gate for removing the legacy shim later.
• Tests — suites updated for the move plus new coverage: heavy-worker boot, legacy-shim forwarding (jobId/opts preservation, malformed-payload fall-through), route-level heavyQueue wiring, heavyJobDataSchema, build-phase fake queue, docker entrypoint roster.

Test plan

• pnpm lint
• check-types for worker, builder, @chatbotx.io/worker-config, @chatbotx.io/business, @chatbotx.io/sdk, @chatbotx.io/integration-whatsapp
• pnpm --filter worker test (194 files, 1841 tests)
• pnpm --filter builder test (312 files, 2027 tests)
• pnpm --filter @chatbotx.io/business test (1527 tests), @chatbotx.io/worker-config (25), @chatbotx.io/integration-whatsapp (215)
• apps/worker tsdown build produces dist/heavy/worker.mjs (docker entrypoint auto-discovers it)
• scripts/check-coexist-drain.mts runs clean against local Redis
• Manual: after deploy, WhatsApp coexist webhook → buffer → flush lands on bull:heavy and history imports complete
• Manual: ~1 day after deploy, run pnpm tsx scripts/check-coexist-drain.mts; when clean, open the follow-up PR removing the legacy shim + script