A Review for Deep Reinforcement Learning in Atari:Benchmarks, Challenges, and Solutions