Zero-Regret Performative Prediction Under Inequality Constraints