[clang] [AMDGPU] Use generic builtins for `wave_reduce` ops (PR #179589)

[Alex Voicu via cfe-commits]

AlexVlx wrote:

> These aren't "generic builtins".
What do you actually mean here. Please use more words. The algorithm itself is generic, that it lowers to a dedicated intrinsic is inconsequential.

> 
> Also the FP and integer operations should absolutely not share a function name. Especially min and max

Why? Again, the algorithm itself is generic, you're just making life slightly harder for the user for no clear benefit, why would a HLL user care that the **intrinsic** that gets selected is reduce_fmin or reduce_fmax?

https://github.com/llvm/llvm-project/pull/179589