An Extension of Fano's Inequality for Characterizing Model Susceptibility to Membership Inference Attacks