Potential and Challenges of Model Editing for Social Debiasing