Mean-based Best Arm Identification in Stochastic Bandits under Reward Contamination