Assessing Disparate Impact of Personalized Interventions: Identifiability and Bounds