[llvm] AMDGPU: Fix vector handling in pown libcall simplification (PR #95832)

[Matt Arsenault via llvm-commits]

arsenm wrote:

* **#95832** <a href="https://app.graphite.dev/github/pr/llvm/llvm-project/95832?utm_source=stack-comment-icon" target="_blank"><img src="https://static.graphite.dev/graphite-32x32-black.png" alt="Graphite" width="10px" height="10px"/></a> 👈
* `main`

This stack of pull requests is managed by Graphite. <a href="https://stacking.dev/?utm_source=stack-comment">Learn more about stacking.</a>
<h2></h2>

 Join @arsenm and the rest of your teammates on <a href="https://graphite.dev?utm-source=stack-comment"><img src="https://static.graphite.dev/graphite-32x32-black.png" alt="Graphite" width="11px" height="11px"/> <b>Graphite</b></a>
  <!-- Current dependencies on/for this PR: -->

https://github.com/llvm/llvm-project/pull/95832