Allow specifying a concrete CUDA arch

Currently we set the wanted CUDA arch (virtual/physical) by setting up VTKm_CUDA_Architecture to the specified Architecture generation (Pascal, Maxwell...). In many cases when we specify the CUDA arch in VTK-m, it will internally enable more than one specific architecture of the same arch generation.

This is normally useful since we can give every bit of optimizations for each GPU pertaining to the same generation. On the other hand it also results in a size increase of the VTK-m binaries.

We need to give the users a way to specify a single CUDA arch for wish they wish their VTK-m build be built with.

cc @cyrush1