The Pin unsoundness and `CoerceUnsized`

[RalfJung]

@withoutboats

I think the main problem here might be with clone: its seems very plausible there exists a combination of Clone on a custom pointer and CoerceUnsized on that pointer that, when combined with the Clone and CoerceUnsized impls on Pin, give you a falsely pinned pointer. But we can't require that Clone return a pointer that points to the same value.

For my argument above to work, we basically have to require that doing the unsizing coercion does not change the Deref/DerefMut/Clone impl that will be called. Because if that changes, there is just no way to show the proposed Pin invariant for the newly coerced pointer.

With that constraint, how could Clone and CoerceUnsized cooperate to cause a problem here?