Allow example tests to be run on subprocesses

[leofang]

Right now we just exec all tests in the main test process https://github.com/NVIDIA/cuda-python/blob/213baf40966d8c449c62cce2c55e945a99c452b1/cuda_core/tests/example_tests/utils.py#L35 which is fine if there is no subtle bugs in the examples. However, it is not the end user experience (ex: from user perspective CUDA contexts are implicitly created/tear down between each sample runs). We should allow configuring it in the future.