[llvm] [AMDGPU] Split wide constant address space stores (PR #226069)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> I'd prefer to treat it as deleting the instruction rather than handling it the same way as a global store, similar to how the compiler treats other UB in the code.

Depends what turns into less effort, I wouldn't put too much work into optimizing this to nothing 

https://github.com/llvm/llvm-project/pull/226069