On the Statistical Accuracy of Biomarker Assays for HIV Incidence