[RFC] Deprecate non-standard constructors in std::pair

[Jonathan Wakely]

This variation on the example in https://wg21.link/lwg811 is supposed
to be ill-formed:

   pair<char*, unique_ptr<int>> p(0, std::make_unique<int>(0));

This cannot use the pair(U1&&, U2&&) constructor because it deduces
int for U1, which isn't convertible to char*.  But it also cannot use
the pair(const first_type&, const second_type&) constructor, because
second_type is move-only.

It works in libstdc++ because extra non-standard constructors were
added by https://gcc.gnu.org/PR40925 when implementing LWG 811:

   template<typename U1, /* constraints */>
   explicit(/*...*/)
   pair(U1&&, const second_type&);

   template<typename U2, /* constraints */>
   explicit(/*...*/)
   pair(const first_type&, U2&&);

I think we should deprecate these extra constructors.

Unless I'm mistaken, those non-standard constructors have two uses:

- Support literal 0 as a constructor argument for a pointer type.
   Such code should use nullptr instead. It exists to solve the problem
   of 0 not deducing as a pointer type.

- Support {} as a constructor argument for a move-only type. I don't
   think this was an intentional design choice (we don't test for it)
   but it does work. Adopting https://wg21.link/p1951 for C++23 will
   make that work anyway.

I propose that we deprecate the constructors for C++11/14/17/20 in
stage 1, and do not support them at all in C++23 mode once P1951 is
supported. I have a patch which I'll send in stage 1 (it also uses
C++20 concepts to simplify std::pair and fix PR 97930).

After a period of deprecation we could remove them, and support P1951
for -std=gnu++11/14/17/20 too so that {} continues to work.