[llvm] [AArch64] Lower llvm.lrint to SVE2p2 FRINT*X instructions (PR #187319)

[Sander de Smalen via llvm-commits]

https://github.com/sdesmalen-arm requested changes to this pull request.

We discussed this offline, so just for the record here; these operations could just use `frintx + fcvtzs` (similar to what happens for NEON) and don't require the special `frint32/64x` instructions. Those instructions are useful to avoid the GPR <-> FPR transfer for something like `(double)(int)double_value` that could use FRINT32Z.

https://github.com/llvm/llvm-project/pull/187319