Towards Skilled Population Curriculum for Multi-Agent Reinforcement Learning