In Fisher's paper, he did not include the error term.
But Durbin & Watson suggested the error term, and also made the matrix form of the modern version of regression.
Finally, I found this sentence in Fisher's paper:
"If errors occur in the heights, this will not influence the regression of height on age, provided that at all ages positive and negative errors are equally frequent, so that they balance in the averages. "
This means Fisher included the error term implicitly and take the average of the equation. So, the error term on average is zero in his paper. Durbin & Watson suggested the error term explicitly.