Provide functionality for response variable transformation

[graysonwhite]

It would be really nice to have flexibility in the linear mixed model to model the response as a function of $y$, but still return estimates with respect to $y$, i.e.,
$f(y) = X\beta + u_j + e_{ij}$,
but return $\hat \mu$ as the average of $y$, not $f(y)$.

Here, we should only have to require that $f$ has an inverse. We will need find and then compute the inverse on each unit the linear mixed model has predicted onto in the estimation and bootstrap MSE estimation processes.

[joshyam-k]

implemented in b7e74ba