Improve documentation on cluster size inference

[nicholst]

Ever since cluster size inference for CIFTI was added (with #166) there has been two possible types of cluster images written out clustere and clusternorm, and this is only really documented in the PR.

Extend the in-code documentation and the web documentation to describe these more thoroughly. Help-list response pasted below for reference.

As part of CIFTI inference we wanted to be able to provide cluster size inferences for 'brain ordinates' data, consisting of the surface and subcortical volumes. However, the null distribution for surface clusters and volume clusters will be totally different. We tried some simple transformations (e.g. square root for surface, cube root for volume) but they didn't work well, and so we instead use a Box-Cox transformation on the null distribution of surface and volume clusters separately. Once we have Box-Cox transformed the cluster sizes, we use the median and the half-quartile (Q3-Q2) of the Box-Cox-transformed values to shift and scale into approximate Z scores. What was once "k_E" (cluster extent) is now "k_Z" (normalised cluster extent).

Hence, if you look at the Results page carefully, you should see "Wild Bootstrap norm. ext. thresh. k_{Z} > ..." indicating that the transformation has happened, and then you'll get swe_clusternorm* files. Except....

Sometimes, when there are very few clusters or they are mostly just tiny, of size 1 or 2, the half-quartile can be 0 and the transformation breaks down. In that case, we throw up our hands and don't do any Box-Cox standardisation and resort to just good old element- and voxel-counting to measure cluster size. Then you'll get "Wild Bootstrap extent threshold k > ..." and swe_clustere* files.