Learning Agile Locomotion and Adaptive Behaviors via RL-augmented MPC