"Good Robot! Now Watch This!": Repurposing Reinforcement Learning for Task-to-Task Transfer