Reinforcement Learning with Elastic Time Steps