better documentation for missing data and EE and EZ model

[stephens999]

when there is a lot of missing data, or the standard errors differ a lot across observations, the EE model would be safer than the EZ (which I think is the default?)

We should probably document this better.

[stephens999]

actully now I think that the EE model is default?

[stephens999]

I believe that our recommended way to deal with missing data is to set b_ij=0, s_ij=infty (or very large).
But does this cause any problems with estimation of null correlations?
What about the ED step?
What about if we use EZ model?

Do all the pipeline components play nicely with this approach to missing data?