FairSIN: Achieving Fairness in Graph Neural Networks through Sensitive Information Neutralization