Three Pathways to Neurosymbolic Reinforcement Learning with Interpretable Model and Policy Networks