Comparison of classical, kernel-based, and nearest neighbors regression estimators using the design-based Monte Carlo approach for two-phase forest inventories