SlateQ agent implementation

[rahul-zomato]

Is next_state deliberate here in next_q_values calculation in slateQ agent - https://github.com/facebookresearch/ReAgent/blob/main/reagent/training/slate_q_trainer.py#L230

SlateQ agent implemented by SlateQ paper authors in recsim uses state instead of next state from replay buffer to get next_q_values - google-research/recsim#26