Apple: Toward General Active Perception via Reinforcement Learning