[PERF]: Converting arguments of generic int or fixed-size int (e.g. cuuint32_t) are suboptimal

[mdboom]

In driver, runtime and nvrtc, the conversion of an argument that accepts any Python int or None, but converts to one of the fixed-size C ints in CUDA (e.g. cuuint32_t) looks like:

 cdef cydriver.cuuint32_t cytensorRank
 if tensorRank is None:
 ptensorRank = 0
 elif isinstance(tensorRank, (cuuint32_t,)):
 ptensorRank = int(tensorRank)
 else:
 ptensorRank = int(cuuint32_t(tensorRank))

This can be reduced to:

 cdef cydriver.cuuint32_t cytensorRank
 if tensorRank is None:
 ptensorRank = 0
 else:
 ptensorRank = int(tensorRank)