Addressing the Plasticity-Stability Dilemma in Reinforcement Learning