[llvm] [AMDGPU] Strengthen preload intrinsics to noundef and nonnull (PR #92801)

[Matt Arsenault via llvm-commits]

https://github.com/arsenm commented:

We do lower the kernarg intrinsic as null in some cases:

https://github.com/llvm/llvm-project/blob/a91d5c07f2357f10a5378bb3b0e439847f2b8e00/llvm/lib/Target/AMDGPU/SIISelLowering.cpp#L2153


While you're touching this, we should also create a way to declare the new range attribute on intrinsics and set them to the maximum. 

https://github.com/llvm/llvm-project/pull/92801