provide numerical evaluation of the 4th spatial derivatives for covariance matrix

[karenyyng]

Thoughts after questions from AmpLab staff Jey Kottalam:
So far we have been providing analytical expressions for the 4th spatial derivatives for the covariance kernel, it is possible for us to explore a numerical evaluation of the 4th spatial derivatives of other kernels through finite differencing if needed.

It is unclear to me which approach will be computationally faster.
Once the numerical differencing infrastructure is set up, it can be more general.

[karenyyng]

Have to consider the possible numerical precision issues - the order of accuracy / no. of terms to include matters. https://en.wikipedia.org/wiki/Finite_difference_coefficient