Is more always better? The impact of vehicular trajectory completeness on car-following model calibration and validation