Interpreting p-Values and Confidence Intervals Using Well-Calibrated Null Preference Priors