Empirical best linear unbiased prediction in cultivar trials using factor-analytic variance-covariance structures