[llvm] [AMDGPU] Sink uniform buffer address offsets into soffset (PR #169230)

[Krzysztof Drewniak via llvm-commits]

krzysz00 wrote:

Update: I asked Claude to do some Z3-age to find the conditions here!

Good news: the check we care about, given the actual hardware behavior, is extremely simple ... we just need all these additions to be `nuw`.

But, given that, we don't need to worry about alignment or anything else I flagged above. 

See attached (and can someone hunt down the old Jira ticket here?) 
[buffer_offsets_proof.py.txt](https://github.com/user-attachments/files/28849666/buffer_offsets_proof.py.txt)
[buffer_offsets_wrapping_proof.py.txt](https://github.com/user-attachments/files/28849667/buffer_offsets_wrapping_proof.py.txt)


https://github.com/llvm/llvm-project/pull/169230