Computably Continuous Reinforcement-Learning Objectives are PAC-learnable