[llvm] [AMDGPU] Document that only naturally aligned atomics of up to 64 bits are supported by the AMDGPU backend (PR #200167)

[Fabian Ritter via llvm-commits]

ritter-x2a wrote:

Ping. The approval was before I moved the size/alignment restriction of atomics to its own "Unsupported IR Constructs" section, therefore I don't see it as sufficient to merge the PR. Please let me know if you have suggestions to improve the wording or the section's location in the doc.

https://github.com/llvm/llvm-project/pull/200167