[llvm] AArch64: Allow ZEXT+COPY -> FMOV peephole for ZPR registers as well (PR #135436)

[Matthias Braun via llvm-commits]

MatzeB wrote:

This avoids an unnecessary `mov w8, w8` in cases like:
```
char* f(char* p, __SVUint32_t s) {
  unsigned a = s[0];
  return p + a;
}
```


https://github.com/llvm/llvm-project/pull/135436