CARL: Controllable Agent with Reinforcement Learning for Quadruped Locomotion