Add stream.sync() in cuda_core/tests/test_launcher.py::test_launch_invalid_values

This is ONLY A BAND-AID, but a very effective one:

Andy's original suggestion:

* NVIDIA/cuda-python-private#245 (comment)

Results of extensive testing:

* NVIDIA/cuda-python-private#245 (comment)

Long-term:

* #1539

Author: rwgk

cuda_core/tests/test_launcher.py

@@ -150,6 +150,7 @@ def test_launch_invalid_values(init_cuda):
         launch(StreamWrapper(stream), config, ker)
 
     launch(stream, config, ker)
+    stream.sync()  # TODO(#1539)
 
 
 # Parametrize: (python_type, cpp_type, init_value)