Implement the ranked automata evaluation to guarantee a O(|D|x|Q|)

worst case complexity. The states of an automaton are given a rank (an
integer) such that if rank(q) = i, then state q can be satisfied
during run i at the latest. The maximum rank of a state is bounded by
the number of "bi-directional" states, that is, the number of states
which depends at the same time on a state upward and on another state
downard. Therefore, the rank is at most |Q|. For the moment, we
evaluate state [q] *only* during run i. This implies the O(|D|x|Q|)
overal complexity but also means we loose a chance to discard (or
accept) [q] early. A heuristic could improve that.