[llvm] [AArch64] Use PNR rather than PPR register class for aarch64svcount (PR #202394)

[Benjamin Maxwell via llvm-commits]

MacDue wrote:

> I suspect this just moves the problem, albeit to one with less of an effect.

In my vectorization tests this seems to solve the issue. There's never really a case where you want to cast/copy directly between the PNR and PPR register class as the resulting masks don't have any useful meaning.  

https://github.com/llvm/llvm-project/pull/202394