Integrating Episodic Memory into a Reinforcement Learning Agent using Reservoir Sampling