[llvm] clastb representation in existing IR, and AArch64 codegen (PR #112738)

[Benjamin Maxwell via llvm-commits]

https://github.com/MacDue approved this pull request.

LGTM, I think @arsenm's main concern as been addressed as there's no new target lowering hook anymore. I think the PR description needs updating though as the `llvm.vector.extract.last.active` intrinsic now exists and this patch is improves the SVE lowering. 

https://github.com/llvm/llvm-project/pull/112738