A Study of Plasticity Loss in On-Policy Deep Reinforcement Learning