If we are expecting a `&Foo` and get a `&something`, when checking the
`something`, we are *expecting* a `Foo`, but we shouldn't try to unify whatever
we get with that expectation, because it could actually be a `&Foo`, and `&&Foo`
coerces to `&Foo`. So this fixes quite a few false type mismatches.
The stand-alone `unify` requires that the type doesn't contain any type
variables. So we can't share the code here for now (without more refactoring)...