Comparison of calibrations based on partial least squares and multiple linear regression for near infrared prediction of composition and functionality in grains