[clang] [Clang] More aggressively mark this* dead_on_return in destructors (PR #183347)

[Florian Hahn via cfe-commits]

fhahn wrote:

> > Not sure if there is anything in the repro that makes it UB that I missed.
> 
> It's the same issue as [steveicarus/iverilog#1311](https://github.com/steveicarus/iverilog/pull/1311). The allocator is writing to the object being deleted within `operator delete`, which as far as I'm aware is not allowed.

But at the point when operator delete gets the pointer passed, the object does not exist anymore, but isn't the  underlying storage is still valid? So in the example, we would not write the object, but to the underlying storage.

Besides writing, couldn't the  allocator still keep around pointers to the underlying storage in general? 

https://github.com/llvm/llvm-project/pull/183347