[clang] [Clang] Reject overflow behavior types on _Atomic (PR #222179)

[Justin Stitt via cfe-commits]

JustinStitt wrote:

@efriedma-quic 
> You could do the overflow computation on the result of atomicrmw, which would be pretty cheap. I can see why you wouldn't want to do that, though; the trap wouldn't trigger until after the result is visible to other threads, which might be unintuitive.


I figured while we're under `-fexperimental` it's better to remove this prickly interaction and we can always go back and improve _Atomic+OBT interaction in the future (although I don't really see that being necessary until we maybe add saturating obt behavior).

https://github.com/llvm/llvm-project/pull/222179