Gllvm update

[BertvanderVeen]

• Adds randomCoefPlot functionality for constrained and concurrent models. Requires calculating prediction errors that require covariance of fixed-effects and variational means. This uses block inversion, but should work if the usual block inversion in gllvm.TMB also works. For the Laplace approximation this requires building the full Hessian of random- and fixed-effects. Does not work with quadratic effects (yet, but I don't plan to support that as it is very difficult to get uncertainties then).
• Slightly adapts the coef function for more user-friendly parameter names and so that it can be subsetted by providing a name
• Updates the vignette with some of these things
• Fixes a minor bug in n.init.max functionality
• Gives the possibility to estimate the Tweedie Power parameter
• Changes default for rotate in summary to TRUE, as in ordiplot
• Changed explicit inversion in standard errors to solving the linear system as it is usually measurably faster. Also added a back-up block inverse via the inverse of the fixed-effects part of the Hessian

[BertvanderVeen]

@JenniNiku this is ready to accept