fix: execute mode cards through deterministic assistant route

tests/integration/test_assistant_capability_execution_matrix.py

@@ -645,7 +645,7 @@ def test_every_assistant_callable_capability_reaches_the_real_tool_boundary(
     assert action_event["success"] is True
 
 
-def test_every_catalog_mode_can_execute_through_the_assistant_tool_boundary(
+def test_every_catalog_mode_executes_through_the_deterministic_assistant_route(
     vehicle_service: SeatDemoService,
     catalog: AssistantCapabilityCatalog,
 ) -> None:
@@ -690,7 +690,7 @@ def test_every_catalog_mode_can_execute_through_the_assistant_tool_boundary(
             return result
 
         client.on_turn = invoke_tool
-        _stream_with_catalog(
+        result, events = _stream_with_catalog(
             assistant,
             catalog,
             message,
@@ -700,9 +700,11 @@ def test_every_catalog_mode_can_execute_through_the_assistant_tool_boundary(
 
         assert raw_results[0]["status"] == "succeeded", mode["id"]
         assert raw_results[0]["matched_modes"][0]["id"] == mode["id"]
-        assert tool_results[0]["verified"] is False
-        assert tool_results[0]["verification_status"] == "simulated"
-        assert tool_results[0]["actions"], mode["id"]
+        assert tool_results == []
+        assert client.turn_calls == 0
+        assert result["route"] == "mode_profile_execute"
+        assert result["success"] is True
+        assert events[0]["engine"] == "n7web-mode-profile-runner"
 
 
 def test_every_catalog_mode_rejects_incompatible_explicit_seat_scopes(