[Developers] new code version

[dave fournier]

I came across some problems with models that have a large (~2800)
number of random effects. the problem is that by default when doing
the estimated covariance matrix the code estimates the covariances
for the random effects as well.   this takes a long time. I added an option

     -noreders

which leaves out the random effects.  The new code will be passed to
Johnoel for whatever.