[clang] [llvm] [clang][AArch64] Emit generic IR instructions instead of `llvm.aarch64.neon.{s,u}{min,max}` (PR #225339)

[Ömer Sinan Ağacan via llvm-commits]

osa1 wrote:

The previous version of this PR (before the "revert" commit) was basically the same as b5a3b8b704cf7f54531f18966cdcc25f5c4e5044, just without the `.td` change which kept it backwards compatible, but as @efriedma-quic pointed out that made the intrinsic untestable from an IR test file. (you'd have to generate it in memory)

I wasn't sure about removing the intrinsic as that will be a breaking change for the LLVM API users. (IR generators will be fine as `AutoUpgrade` will deal with it)

If that's not a concern I can

1. Revert to the previous version of this PR.
2. Update the `.td` to remove the intrinsics entirely.

https://github.com/llvm/llvm-project/pull/225339