Credit Assignment as a Proxy for Transfer in Reinforcement Learning