A Simple Open-Loop Baseline for Reinforcement Learning Locomotion Tasks