The Sliding Regret in Stochastic Bandits: Discriminating Index and Randomized Policies