Verifiable Feature Attributions: A Bridge between Post Hoc Explainability and Inherent Interpretability