[R-pkgs] New version of glmmML

[Göran Broström]

A new version, 0.65-1, of glmmML is now on CRAN. It is a major rewrite
of the inner structures, so frequent updates (bug fixes) may be
expected for some time.

News:

* The Laplace and adaptive Gauss-Hermite approximations to the log
likelihood function are fully implemented. The Laplace method is made
the default. It should give results you can compare to the results
from 'lmer' (for the models that glmmML can handle).

* Binomial responses can now be represented as a two-column matrix
with No. of successes and No. of failures, respectively, as in glm.

* New functions: 'ghq' for calculating the constants used in the
Gauss-Hermite quadrature. 'extractAIC.glmmML', which makes it possible
to use functions like 'dropterm' (MASS) on glmmML fits.

* There are three choices of distribution for the random effects:
'gaussian' (default), 'logistic', and 'cauchy'.

* The 'conditional' bootstrap is removed, so the only choice now is
the parametric bootstrap.

* The 'posterior.means' are no longer calculated, leaving only the
'posterior.modes' as 'predictions' of the random effects.

As usual, feedback is more than welcome!

Göran