previous implementation was not thread/stream safe, since it used constant
memory

new implementation doesn't use any global objects, so it is thread/stream safe

Add GPU models in seperate folder, only old models are supported and they were replaced to make the CPU interface work.
This solves the undesired consequence of breaking the GPU interface.

move hog::set_up_constants from constructor to compute method

if user changed CUDA device between constructor and computation,
some variables were uninitialized

The architecture has been correctly determined earlier in this
file. No need to do it again.

MinGW-w64 always uses "w64" as vendor key which the previously
used check for "64" anywhere in the target triplet matched. This
would lead to MinGW-w64 setups always being treated as x64.

Since we are actually interested in finding the "64" in the
triplet's operating system key, this fix makes sure to look for
"mingw64" and thus avoids the false match.

Fixes issue #3835. Credits to Lode Leroy for the original patch.