Contrastive Tuning: A Little Help to Make Masked Autoencoders Forget