The MNL-Bandit Problem: Theory and Applications