lift missings in response, fitted, residuals? #650

Currently, data rows that containmissings in the model inputs are silently dropped, so the output vector is shorter than the input by however many rows had missings. I think it would be more user-friendly to lift missings and return a vector with the same number of elements as the input has rows, with missings at indices that had missings in the inputs.