[clang] [Coroutines] Allow [[clang::coro_wrapper]] for class (PR #93268)

[Chuanqi Xu via cfe-commits]

ChuanqiXu9 wrote:

> @ChuanqiXu9, I just ran across this PR -- and I immediately wondered if it could be used to improve stack-use-after-free checks for `folly::result` (the short-circuiting coro you saw on my recent PR).
> 
> Like the classes you mention in the PR description here, `folly::result` has many reasonable "coro" and "coro wrapper" uses, and it would be onerous to annotate every wrapper usage as `coro_wrapper`.
> 
> What did you envision the effect of adding `coro_wrapper` to a class would be on lifetime checks?
> 
> Can you walk me through your reasoning / expected usage?

I think it makes sense to add  `coro_wrapper` to classes. The reasoning is, the behavior of coroutine is controlled by its promise_type. And generally in practice, we have the consistent promise_type for the same return type, which is the class I mentioned here. So given the coroutines have the same behavior, it should be helpful to add these attributes to the class. And of course, as I said, it is best to add this to the promise_type, but it is not conflicting.

https://github.com/llvm/llvm-project/pull/93268