[clang] [Clang][AMDGPU] Add __builtin_amdgcn_s_buffer_load_<T> (PR #203352)

[Krzysztof Drewniak via cfe-commits]

krzysz00 wrote:

While this isn't a hard objection, we really should be moving away from this intrinsic, both on account of it taking a `<4 x i32>` (breaking alias analysis) and of it being marked as not affecting memory even though it does (proving the hoistability of a uniform buffer load should be LLVM's job)

https://github.com/llvm/llvm-project/pull/203352