[clang] [CUDA] Implement __CUDA_ARCH_LIST__ macro and refactor architecture helpers (PR #175260)

[Artem Belevich via cfe-commits]

Artem-B wrote:

> This seems reasonable, I guess the refactor I did earlier to move the architecture list from the Toolchain to the Driver made this pretty easy

It did indeed turn out to be much less hairy than my attempt few years back. Thank you!

https://github.com/llvm/llvm-project/pull/175260