store the last game state for end_of_round

[andquintero]

When running the end_of_round function, the last_game_state was actually the previous game state, this stores the real final game state

[fdraxler]

Thanks for the PR. To avoid off-by-one confusion, can you specify which state you expected? My idea was that the last game state is None (i.e. the state after the game ended), so I tried to pass the state the agent last took an action as last_game_state to end_of_round.