Update libcudacxx NVRTC testing to emit SASS instead of PTX

[jrhemstad]

The libcudacxx NVRTC testing mode currently is only configured to emit PTX.

This is problematic as it relies on PTX JIT from the driver in order to run. However, if you compile with a newer version of NVRTC than the driver supports, this will fail.

Instead, we should update the NVRTC config to be able to emit SASS and bypass the need for PTX JIT, allowing it to run on older drivers.

Example from CTK docs on how to emit SASS: https://docs.nvidia.com/cuda/cuda-c-best-practices-guide/index.html#dynamic-code-generation