[llvm] [AMDGPU] Make sort ordering in `AMDGPURewriteAGPRCopyMFMAImpl::eliminateSpillsOfReassignedVGPRs()` strict weak. (PR #162493)

[Matt Arsenault via llvm-commits]

================
@@ -490,7 +490,9 @@ void AMDGPURewriteAGPRCopyMFMAImpl::eliminateSpillsOfReassignedVGPRs() const {
       return true;
 
     // Tie breaker by number to avoid need for stable sort
-    return A->reg().stackSlotIndex() < B->reg().stackSlotIndex();
+    // The ordering have to be strictly weak.
----------------
arsenm wrote:

```suggestion
    // The ordering has to be strictly weak.
```

https://github.com/llvm/llvm-project/pull/162493