feature request: posterior_epred method for logistic_normal

[amynang]

Currently, this results in an informative error.

Is there a technical difficulty in implementing this for this particular family or is it a matter of priority?

[paul-buerkner]

It's technical since the mean of the logistic normal distribution is non-analytic. We could use a short-cut of implementing the median rather the mean. This may be inconsistent with other families but would still provide sensible results. I will consider it.