[SHTns] up to v3.7 and add CUDA support

[fgerick]

No description provided.

[fgerick]

Does someone know what has to be done to keep the cuda builds from failing? When I run this locally it works fine and it seems it can build the library just fine, but then the dlopen checks fail. When using dont_dlopen=true the builds end up in an infinite loop.

[fgerick]

Okay, after quite some painful trial and error I believe this is finally working. I've removed the micro-architecture expansion, that I think is not necessary with the removal of the -mtune=skylake flag that was still around in the build.

I hope this is correct how the CUDA and non-CUDA versions are build now and that the x86_64-linux-gnu-cuda+none is the correct way to do it. I do not manage to build everything without this flag for the CPU-only version.

[fgerick]

Who should I get in touch with to approve this? Thank you.