Confident Off-Policy Evaluation and Selection through Self-Normalized Importance Weighting