[clang] [llvm] [ARM] Fix operand order for MVE predicated VFMAS (PR #115908)

[Simon Tatham via cfe-commits]

https://github.com/statham-arm approved this pull request.

LGTM. I suppose the error wouldn't have shown up if you did the simplest possible end-to-end test along the lines of
```c
float32x4_t foo(float32x4_t a, float32x4_t b, float32_t c, mve_pred16_t p) {
  return vfmasq_m(a, b, c, p);
}
```
but only shows up when more complicated register allocation starts happening around it!

https://github.com/llvm/llvm-project/pull/115908