fix: address code quality issues in session pool, approval, and main

- session.ts: add onReconnect callback so reconnected socket replaces stale pool entry
- session-pool.ts: pass onReconnect to update pool on reconnect; add closeAll() to gracefully end WebSocket connections on shutdown
- approval.ts: use approved flag so double-approval race (updateMany count=0) returns null instead of sending duplicate forward jobs
- main.ts: wrap pool.add() in try/catch to continue if one account fails; replace silent groupMap fallback with explicit error log and early return; call pool.closeAll() before BullMQ shutdown to prevent hanging sockets
- Tests: add closeAll() test to session-pool.test.ts; add double-approval race test to approval.test.ts (56 tests total, all passing)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

apps/worker/src/core/approval.ts

@@ -30,12 +30,14 @@ export async function handleStarReaction(
   if (message.status !== 'PENDING') return null;
   if (message.approval) return null;
 
+  let approved = false;
   await prisma.$transaction(async (tx: any) => {
     const updated = await tx.message.updateMany({
       where: { id: message.id, status: 'PENDING' },
       data: { status: 'APPROVED' },
     });
     if (updated.count === 0) return; // another admin approved first — idempotent
+    approved = true;
     await tx.approval.create({
       data: {
         messageId: message.id,
@@ -45,6 +47,8 @@ export async function handleStarReaction(
     });
   });
 
+  if (!approved) return null;
+
   const jobs: ForwardJobData[] = message.sourceGroup.syncRoutesFrom
     .filter((route: any) => route.targetGroup != null)
     .map((route: any) => ({