Bias- and efficiency-robustness of general M-estimators for regression with random carriers