Lexicographic probabilities and iterated admissibility