Scalable Rule-Based Representation Learning for Interpretable Classification