[Bug/Documentation] Old and new game states report same step

[OoJJBoO]

We noticed that the old_game_state and new_game_state dictionaries that are passed to the game_events_occurred agent callbacks contain the same "step" value.

This seems to not always having been the case, as we found that this behavior was first introduced in commit 35b4965. The issue is caused by the agent state being stored here after the step was already incremented here.

Is this intended behavior? If so, we would suggest updating the documentation to communicate this more clearly to prevent misunderstandings in the future.

For instance, our team tried to use the stored step to retrieve cached data, which of course did not work as we had intended because of the way step reporting is handled at the moment.