* add more dqn benchmark result; unify train scripts

* resize benchmark picture

* resize benchmark picture, refine comments of args

* change dependence, mujoco only support python3 now