[clang] [CIR] Handle return with cleanups (PR #163849)

[Andy Kaylor via cfe-commits]

andykaylor wrote:

> > That would simplify the EHScopeStck handling significantly in CIR, but the complexity would just be moved to the FlattenCFG pass, and this would mean we were diverging from the classic codegen code structure.
> 
> I'm up for the tradeoff here, the lower level alternative will make code way harder to analyze.

I'm not certain which tradeoff you're talking about. Are you saying we should proceed with representing the `EHStackScope` in CIR? Or that we should stick with the OG-like branching through the cleanups? If we go with representing the `EHStackScope` in CIR, a lot of the code in this PR will not be needed, and so in that case I might be inclined to not commit this one at all and wait for the new cleanup representation.

https://github.com/llvm/llvm-project/pull/163849