PLASTIC: Improving Input and Label Plasticity for Sample Efficient Reinforcement Learning