[clang] [Clang] Handle references to concept in Subst* nodes (PR #199671)

[Corentin Jabot via cfe-commits]

cor3ntin wrote:

> Why doesn't this break the 'deferred instantiation/checking' rules? My understanding is that we're not allowed to instantiate these until we need them for checking, since they are allowed to Change up until the first time they are checked.

The thing we should not do here - and I need to check we are not doing the wrong thing is failing when the substitution in the concept arguments - which might not be used in the atomic constraint - fails,

I guess there is a question of what happens if there is an error in a non sfinae context in the concept arguments.

https://github.com/llvm/llvm-project/pull/199671