How Reinforcement-Learned Teachers (RLTs) use dense rewards and a focus on explanation to unlock latent teaching skills in small models, creating a more efficient path to powerful AI