Improve speed of SECNice

[odunbar]

Purpose

Currently NICE performs an entry based calculation of correlations, scaling with number of matrix entries in an input x output and output x output sized matrix. For larger systems this can become expensive.

Possible solution

Replace the current entry-based calculation by a precomputed (or compute-once) interpolator of [-1,1] for large systems. i.e. we compute for fine resolution of [-1,1] forming an interpolation object that is then used to emulate this calculation in operation.

[odunbar]

We resolved this in #367