fix(runner): fix audio task infinite loop and semaphore race

Two bugs caused audio tasks to loop endlessly:

1. claimRecord never set heartbeat_at — newly claimed tasks had
   heartbeat_at=null, which matched the reaper's stale filter
   (heartbeat_at=null || heartbeat_at<threshold). Tasks were reaped
   and reset to pending within seconds of being claimed, before the
   30s heartbeat goroutine had a chance to write a timestamp.
   Fix: set heartbeat_at=now() in claimRecord alongside status=running.

2. Audio semaphore was checked AFTER claiming the task. When the
   semaphore was full the select/break only broke the inner select,
   not the for loop — the code fell through and launched an uncapped
   goroutine that blocked forever on <-audioSem drain. The task also
   stayed status=running with no heartbeat, feeding bug #1.
   Fix: pre-acquire a semaphore slot BEFORE claiming the task; release
   it immediately if the queue is empty or claim fails.

backend/internal/runner/runner.go

@@ -248,23 +248,30 @@ func (r *Runner) poll(ctx context.Context, scrapeSem, audioSem chan struct{}, wg
 	}
 
 	// ── Audio tasks ───────────────────────────────────────────────────────
+	// Only claim tasks when there is a free slot in the semaphore.
+	// This avoids the old bug where we claimed (status→running) a task and
+	// then couldn't dispatch it, leaving it orphaned until the reaper fired.
+audioLoop:
 	for {
 		if ctx.Err() != nil {
 			return
 		}
+		// Check capacity before claiming to avoid orphaning tasks.
+		select {
+		case audioSem <- struct{}{}:
+			// Slot acquired — proceed to claim a task.
+		default:
+			// All slots busy; leave remaining pending tasks for next tick.
+			break audioLoop
+		}
 		task, ok, err := r.deps.Consumer.ClaimNextAudioTask(ctx, r.cfg.WorkerID)
 		if err != nil {
+			<-audioSem // release the pre-acquired slot
 			r.deps.Log.Error("runner: ClaimNextAudioTask failed", "err", err)
 			break
 		}
 		if !ok {
-			break
-		}
-		select {
-		case audioSem <- struct{}{}:
-		default:
-			r.deps.Log.Warn("runner: audio semaphore full, will retry next tick",
-				"task_id", task.ID)
+			<-audioSem // release the pre-acquired slot; queue empty
 			break
 		}
 		r.tasksRunning.Add(1)

backend/internal/storage/pocketbase.go

@@ -247,8 +247,9 @@ func (c *pbClient) claimRecord(ctx context.Context, collection, workerID string,
 	}
 
 	claim := map[string]any{
-		"status":    string(domain.TaskStatusRunning),
-		"worker_id": workerID,
+		"status":       string(domain.TaskStatusRunning),
+		"worker_id":    workerID,
+		"heartbeat_at": time.Now().UTC().Format(time.RFC3339),
 	}
 	for k, v := range extraClaim {
 		claim[k] = v