Many-Objective Reinforcement Learning for Online Testing of DNN-Enabled Systems