From [1], transfer functions evaluated at s=0 are NaN when there are both zeros and poles at the origin:

                    else:
                        # numerator is zero too: give up
                        gain[i][j] = np.nan

We could instead check for an excess of zeros (gain 0), of poles (gain +inf or -inf), or equal number of poles and zeros (gain equal to ratio of lowest non-zero coefficients of numerator and denominator: this is the limit of the TF as s->0).

[1]