Regret Minimization with Performative Feedback