What Makes Treatment Effects Identifiable? Characterizations and Estimators Beyond Unconfoundedness