Tests

uses current CUDAStream correctly (#118) #51

Job	Run time
unittests-linux (3.11, linux.12xlarge, cpu) / linux-job	4m 38s
unittests-linux (3.13, linux.g5.4xlarge.nvidia.gpu, cuda, 12.4) / linux-job	10m 21s
	14m 59s

Provide feedback