Margin Matching Preference Optimization: Enhanced Model Alignment with Granular Feedback