[llvm] [Attributor] Skip AS specialization for volatile memory instructions (PR #107250)

[Matt Arsenault via llvm-commits]

================
@@ -5,11 +5,22 @@ target triple = "amdgcn-amd-amdhsa"
 
 ; HSA-LABEL: {{^}}use_group_to_flat_addrspacecast:
----------------
arsenm wrote:

I don't know why this test is running amdgpu-attributor but it probably shouldn't be? This ought to just be testing the codegen of addrspacecast, optimizing out cases just makes it more confusing 

Anyway this is unbreaking the test since the cast now remains 

https://github.com/llvm/llvm-project/pull/107250