Use different rewards for intermediate and final steps

[WardLT]

Will require modifying the environment class, as rewards do not know the position in an episode.