[llvm] [AMDGPU] Prevent hang in SIFoldOperands (PR #82099)

[Jay Foad via llvm-commits]

jayfoad wrote:

Thanks for the test case! Now can you please try to explain what goes wrong in more detail? I took a very quick look and it seems like this loop never terminates:
```
    for (auto &RSUse : make_early_inc_range(MRI->use_nodbg_operands(RegSeqDstReg)))
```
But why? If we understand the problem better, there may be a simpler fix.

https://github.com/llvm/llvm-project/pull/82099