ProbLog4Fairness: A Neurosymbolic Approach to Modeling and Mitigating Bias