Remove #[rustc_reservation_impl]

[WaffleLapkin]

Based on #160705.

There was only ever one reservation impl and it got (/is being) removed in #160705. I don't think there is a reason to keep the machinery, given how specific and tricky the semantics of it are (see #64631).

Closes #64631.

cc @clarfonthey since you seem to have found another use for the attribute (#64631 (comment)) (I haven't looked into it, so not sure if it might be worth keeping the attribute for that use case).

r? types

[clarfonthey]

The use case I described for rustc_reservation_impl seems relatively different from the semantics being applied for the never reservation. I don't think they'd share too much in common since the intent would be to allow reserving impls in, e.g. core to later be implemented in std as a way of making incoherent impls for traits that don't completely bork the solver.

I think it's fine to remove this specific hack and later replace it with something else if it happens to use the same name.