[llvm] [NVPTX] Support native 64-bit atomic add/sub pre-SM32 (PR #222471)

[Yonah Goldberg via llvm-commits]

YonahGoldberg wrote:

Merging with bypass CI because it was passing before I merged in main and now I'm getting a random AMDGPU test failing. All tests pass except for that AMDGPU test.

https://github.com/llvm/llvm-project/pull/222471