Position-Based Multiple-Play Bandits with Thompson Sampling