Honor of Kings Arena: an Environment for Generalization in Competitive Reinforcement Learning