The Formalism-Implementation Gap in Reinforcement Learning Research