`<memory>`: `uninitialized_meow` should handle `voidify`'s wording

[StephanTLavavej]

WG21-N4885 [specialized.algorithms.general]/4 says that the uninitialized_meow family of algorithms is powered by an exposition-only voidify helper, which has the interesting property of returning modifiable void*. Surprisingly, this intentionally handles destinations that are iterator-to-const, allowing const elements to be constructed in uninitialized memory.

We should add test coverage for this scenario, for all of the uninitialized_meow algorithms that are affected. This may require reworking internal machinery like _Ptr_copy_cat/_Ptr_move_cat. Finally, we should ensure that we don't disrupt the plain meow algorithms, which should not be allowed to modify through iterator-to-const (as they aren't specified with voidify).

Noticed while reviewing #1772, but filed as a separate issue as this is pre-existing, non-trivial, and largely unrelated to that PR's scope.

[frederick-vs-ja]

This is probably invalidated by LWG-3870.

[frederick-vs-ja]

A bunch of test errors in #3475 suggested that test coverage is probably present at some time. But we need to revert it.