[llvm] [AMDGPU] Add target feature require-naturally-aligned-buffer-access (PR #115479)

[Piotr Sobczak via llvm-commits]

piotrAMD wrote:

Can this be merged?
The only unaddressed point is potentially inverting the feature requiring naturally aligned accesses by default, but I do not think that is necessary. As far as I know no API mandate the specific oob behavior (apart from Vulkan VK_EXT_robustness2 mentioned earlier) and do not want to penalize them, as requiring stricter alignment by default will likely produce suboptimal code.

https://github.com/llvm/llvm-project/pull/115479