[Feature Request] Support options for NVBLAS

[mwt]

There is no simple way to get NVBLAS a CUDA accelerated BLAS library working with Windows. It would be pretty cool if R Open could be compiled against this library out of the box as an available option.

It's not difficult to get the library working in Linux, and it seems to provide significant speedups for the multiplication of two large matrices. The library forwards tasks to the CPU BLAS when it cannot speed them up using the GPU.

There is a guide for Linux users here:
https://clint.id.au/?p=1900