Add Matérn covariance function

[kkmann]

Summary

'mmrm' already supports 'spatial' covariance structures. These are essentially covariance functions and allow fitting Gaussian process models.
GP models are particularly useful when considering many observations per group and if it is plausible that there is a temporally decaying correlation between observations. The 'spatial exponential' covariance function is not particularly flexible. In the literature on Gaussian processes, the Matérn class of covariance functions is often used. It can accomodate both smooth and non-smooth sample paths. (see additional information below)

Additional Information

https://andrewcharlesjones.github.io/journal/matern-kernels.html

[danielinteractive]

Thanks @kkmann ! So you started working on this already? Just wondering if it would be good to first accomplish the other methods tasks? :-)

[kkmann]

I think we should put this on hold until the covariance structure implementation is clarified, see #254