Reviving The Classics: Active Reward Modeling in Large Language Model Alignment