As per discussion on IRC. I am about to file an RFC for further
discussion about the more general issue of whether to enforce
invariants through types, typestate, or dynamic checks, but for now,
removing the misleading name "last_unsafe".