Optimal Sequential Decision-Making in Geosteering: A Reinforcement Learning Approach