server: commit operations before enqueueing and answer seq_no races with PREV_HASH_MISMATCH

The operation and receipt now commit before the queue send. A Postgres
23505 on operations_agent_id_seq_no_key is answered with 400
PREV_HASH_MISMATCH carrying the stored chain hash so hook retries can
resync. A failed enqueue after the commit is logged as
operation.enqueue_failed and the receipt is still returned.

Every producer now names its message: MessageQueue.send takes the
durable message id, and the BullMQ adapter passes it as the job id, so
the receipt's queue_message_id is the real queue identifier.

Claude-Session: https://claude.ai/code/session_01QoUjdc7HGKTsgj5GhG8HvF

packages/server/src/scheduler.ts

@@ -12,6 +12,7 @@ import { RedisCacheAdapter } from './adapters/redis-cache.js';
 import { BullMQAdapter } from './adapters/bullmq.js';
 import { createEpoch } from './services/epoch-service.js';
 import { DEFAULT_EPOCH_INTERVAL_MS } from './shared/index.js';
+import { generateUUIDv7 } from './utils/uuid.js';
 
 // ---------------------------------------------------------------------------
 // Initialize adapters
@@ -65,7 +66,7 @@ async function runEpochCreation(): Promise<void> {
 
       if (epoch) {
         try {
-          await queue.send({
+          await queue.send(generateUUIDv7(), {
             type: 'tsa',
             epoch_id: epoch.epoch_id,
             org_id: epoch.org_id,