Verifying Preference Separability for Additive Value Functions