Learning Skateboarding for Humanoid Robots through Massively Parallel Reinforcement Learning