Abstract Interpretation of Fixpoint Iterators with Applications to Neural Networks