Add info about agent speed
env.step()
should return list of agent speeds. this does not need to be computed at each step, but if not costly we should return it at each step. Might be good to implement this already now as later versions of flatland might have varying speeds during episodes.