[OSPP-Week3] fix the problems in the NFSP implementation

[peterchen96]

Thanks to @findmyway for pointing out the problems existing in the current NFSP implementation. I'll list them as the following and fix them sequentially. #375

• use reservoir_trajetory to collect data for sl_agent: Supplement functions in ReservoirTrajectory and BehaviorCloningPolicy #390

• CircularArrayBuffer may not suitable for sl_agent, and I should use the reservoir_trajectory which will randomly replace an old with a new element when the buffer capacity is full.

• replace average_learner with BehaviourCloningPolicy: Supplement functions in ReservoirTrajectory and BehaviorCloningPolicy #390

• average_learner looks similar to BehaviorCloningPolicy. Also, sl_agent just needs to collect states and actions rather than SARTS. Maybe I just need to supplement some BehaviorCloningPolicy's functions is enough to use for sl_agent.

• other problems about the convenience of reusing: Implementation of NFSP and NFSP_KuhnPoker experiment #402 (in progress)

• the work of state encodes can move to the specific env file.

  • encode the state by using its index in the state_space.

• check the run function for NFSPAgentManager, including assertations about the available environment.

  • check the game whether is Multiagent and Imperfect for now.

• modify the experiment file, including design a suitable hook and correct format errors.

  • redesign the hook(ResultNEpisode) to record the result.
  • however, remain some error in the experiment file.