Do high rates of empirical treatment undermine the potential effect of new diagnostic tests for tuberculosis in high-burden settings?