Implement LWG-4242 ranges::distance does not work with volatile iterators

[frederick-vs-ja]

Per LWG-4242/WG21-P3742R0 (alternative link).

Fixes #5623.

This PR makes the involved overload of ranges::distance more constrained than explicitly written, because the "Effects: Equivalent to:" in [range.iter.op.distance]/3 isn't changed and we should propagate the implicit constraints per [structure.specifications]/4.

Drive-by change:

• Removes the skipping for VSO-1898890 (EDGcpfe/26576 EDG produces an error for conversion from array to const reference to pointer) because ranges::distance no longer performs such a conversion.

[frederick-vs-ja]

The constraint sized_sentinel_for<decay_t<_It>> seems wrong for volatile iterators. IIUC even when sized_sentinel_for<_Se, decay_t<_It>> is modeled, correct behavior for volatile version of _It is not yet required.

Also, the const for _Se becomes questionable when volatile-qualified _It comes into the picture - as sized_sentinel_for<_Se, decay_t<_It>> isn't requiring anything about volatile-qualified _It.

I guess it's better to replace decay_t<_It> with another type which preserves volatile for non-array _It. CC @hewillk.

[hewillk]

I'm not sure why you introduced _additionally_subtractable_for_distance. Could you provide an actual example of it having an effect?

[frederick-vs-ja]

I'm not sure why you introduced _additionally_subtractable_for_distance. Could you provide an actual example of it having an effect?

See the test case - as the explicit constraints uses decay_t, they would be satisfied by volatile reverse_iterator<int*>& and reverse_iterator<int*>. Without _Additionally_subtractable_for_distance there will still be hard error. I believe "Effects: Equivalent to:" doesn't allow such a hard error and actually requires the implementation to add additional constraints.

Godbolt link

[hewillk]

Looks like a LWG to me.