Mapping out the Space of Human Feedback for Reinforcement Learning: A Conceptual Framework