LLVM lowering for `cmpxchg` might use different alignment from OG

`cmpxchg` OG codegen gets 8 bytes versus 4 bytes through CIR, see https://godbolt.org/z/TbM8ToPWc

```
// Through CIR
%14 = cmpxchg weak ptr %9, i32 %12, i32 %13 seq_cst seq_cst, align 4
```
vs
```
// Traditional LLVM codegen
%14 = cmpxchg weak ptr %9, i32 %12, i32 %13 seq_cst seq_cst, align 8
```