When instantiating a class template, if the primary class template special member function declaration was declared constexpr or consteval, the instantiation is also considered to be a consteval function even if the instantiation would not be valid to use in a constant expression (http://eel.is/c++draft/dcl.constexpr#7). However, when instantiating such a class, we were incorrectly saying the special member was not a constexpr function when a base class or member variable would cause the instantiation to not be usable in a constant expression. This addresses the issue by falling back to the primary declaration of the explicitly defaulted special member function.
This caused some unexpected test failures with friend declaration checking, but the standard is quite unclear on what's expected there. We also have existing bugs in that particular area (https://godbolt.org/z/cPKGvx746). I raised some questions on the Core reflectors about what is expected and put a FIXME into the test to make it clear that's unresolved.
Incidentally: this was fixed in DR1361, a long time ago. We can remove this comment.