[clang] [clang][msvc] Fix crash when mangling concept template arguments (PR #192951)

[A. Jiang via cfe-commits]

frederick-vs-ja wrote:

> At this point, I’m not entirely sure what the correct direction is for Clang to align with MSVC’s behavior. If the intended behavior is to reject this construct under the MSVC target, that would suggest adding an earlier diagnostic rather than fixing the mangling path. Otherwise, if we keep the current frontend behavior, the mangler would need to handle this case properly.

MSVC implementors probably have to start implementing C++26 concept/variable template arguments ([P2841R7](https://wg21.link/p2841r7)) several months or even more than one year from now. I think we should try to inform them to decide the mangling mechanism for C++26 concept/variable template arguments, or request them to follow Clang's strategy.

At this moment, perhaps we can emit a warning that the current ABI for concept/variable template arguments is not guaranteed to be consistent with MSVC. I believe we shouldn't reject this construct.

https://github.com/llvm/llvm-project/pull/192951