[clang] [AMDGPU] Add clang builtin for generic AMDGPU shuffle (PR #185302)

[Yaxun Liu via cfe-commits]

yxsamliu wrote:

I think keeping it i32-only makes sense since the underlying instructions only work with 32-bit values — users can split wider types themselves without much trouble. We had a similar discussion on #153501 and I tend to agree with the conclusion there. @arsenm what do you think?

https://github.com/llvm/llvm-project/pull/185302