GLS gradient uses WLS.V, which doesn't scale

[yrosseel]

When estimator is GLS (or WLS), the gradient uses (line 257 in lav_model_gradient):
group.dx <- -1 * crossprod( Delta[[g]], crossprod(WLS.V, diff) )
This becomes increasingly problematic if the number of variables goes up (say, for P=200 or higher). For WLS, there is not much we can do, but for GLS, we should use the 'Omega' approach as for ML.