[clang] [llvm] [NVPTX] Fix scoped atomic when given runtime values (PR #185883)

[Joseph Huber via cfe-commits]

jhuber6 wrote:

Ah, well the issue is that if you have something like `__scoped_atoimc_thread_fence(ord, scope)` you will get a runtime expansion of every possible one. The fences and atomics all take the same orders, so those will expand there as well. I suppose for this we could either expand the scope like I'm proposing for clusters here, or just make it a no-op.

https://github.com/llvm/llvm-project/pull/185883