Scientific Discovery and the Cost of Measurement -- Balancing Information and Cost in Reinforcement Learning