hotfix: autoscaler queue_depth NULL depends_on blinded scaler

queue_depth turned a NULL depends_on into [None], and
ready_tasks then required None to be in completed_ids, so
such tasks never counted as ready. Worker pick_next was
unaffected (_task_from_row maps NULL to []), which is why
tasks still ran while the scaler always saw depth 0 and
never scaled up (proven live: depth probe with 1 queued
task returned 0). NULL now means no dependencies, with a
regression test.

cloud/tests/test_autoscaler.py

@@ -85,6 +85,24 @@ class QueueDepthTestCase(unittest.TestCase):
         _set_status(self.db_path, "A", "completed")
         self.assertEqual(queue_depth(self.db_path), 1)
 
+    def test_null_depends_on_counts_as_ready(self):
+        # 回归：depends_on 为 NULL 时必须算无依赖，不能变成
+        # [None] 导致任务永远不算就绪（scaler 致盲）。
+        conn = sqlite3.connect(self.db_path)
+        try:
+            now = time.time()
+            conn.execute(
+                "INSERT INTO tasks(id, project_id, prompt, status,"
+                " idempotency_key, depends_on, priority, created_by,"
+                " created_at, updated_at, last_heartbeat)"
+                " VALUES('N', 'p1', 'x', 'queued', NULL, NULL, 0,"
+                " 'tester', ?, ?, NULL)",
+                (now, now))
+            conn.commit()
+        finally:
+            conn.close()
+        self.assertEqual(queue_depth(self.db_path), 1)
+
     def test_dependency_cycle_counts_zero(self):
         _insert_task(self.db_path, "A", depends_on='["B"]')
         _insert_task(self.db_path, "B", depends_on='["A"]')