Enhancing Model-Based Step Adaptation for Push Recovery through Reinforcement Learning of Step Timing and Region