ChessGPT: Bridging Policy Learning and Language Modeling