Random Shuffling and Resets for the Non-stationary Stochastic Bandit Problem