test(runs): 取消测试扩至 30 case 适配并发执行

case 并行（默认 3）后原 5 case 场景在取消请求到达前就跑完，
取消语义不再被观测到。扩到 30 case（10 波）保证取消窗口充分。

tests/integration/test_runs_api.py

@@ -135,13 +135,15 @@ async def test_start_run_then_get(client, seeded_db, mock_channel):
 
 async def test_cancel_run(client, seeded_db, mock_channel):
     # Seed a multi-case scenario so the run takes long enough to cancel.
+    # Cases run 3-concurrent, so 30 cases ≈ 10 waves × reply_delay — comfortably
+    # longer than the pre-cancel window below.
     from agenteval.models import Case, CaseType, Scenario
     from agenteval.storage.repository import ScenarioRepository
     multi = Scenario(
         id="s-long", name="long-scenario",
         cases=[
             Case(id=f"lc{i}", type=CaseType.SINGLE, messages=[f"m{i}"])
-            for i in range(5)
+            for i in range(30)
         ],
     )
     ScenarioRepository(seeded_db).create(multi)
@@ -170,8 +172,8 @@ async def test_cancel_run(client, seeded_db, mock_channel):
     final = (await client.get(f"/api/runs/{run_id}")).json()
     assert final["status"] == "failed"
     assert final["summary"]["error"]["code"] == "cancelled_by_user"
-    # Not all 5 cases should have run.
-    assert mock_channel.send_calls < 5
+    # Not all 30 cases should have run.
+    assert mock_channel.send_calls < 30
 
 
 async def test_get_run_logs(client, seeded_db, mock_channel):