Fisher-Guided Selective Forgetting: Mitigating The Primacy Bias in Deep Reinforcement Learning