Multivariate Gaussian null

[biona001]

My code for fitting the null of the multivariate model fails when at least 1 component is Gaussian. I believe this is because of a typo in the gradient of loglikelihood wrt to variance component parameters. The (hopefully) correct formula is shown below.

Previously the tau term was missing in the box.

I'm opening an issue since I won't work on the multivariate code soon and I don't want to forget this.