Improve CUDA extension code

[rolfv]

Need to make at least two more changes to the extension code.

1. Have them built unconditionally so they are always available.
2. Have them then properly set the MPIX_CUDA_AWARE_MACRO to 0 or 1 depending on whether library has CUDA-aware configured in. Also have the MPIX_Query_cuda_support() function do the right thing.

The problem is now they are only set when we have CUDA-aware so to use them requires a bunch of ifdefs.

[rolfv]

Duplicating bug that I put in wrong place.

[rolfv]

This is now in master and v2.0.0. Closinig.