Possibility to speed up the computation of F-scores of rank 1 for the Wild Bootstrap

[BryanGuillaume]

In the most recents versions of swe_cp_wb, all F-scores are computed using a for loop over all voxels. This seems needed for F-contrasts of rank > 1 because we need to use the right-matrix division (/) at every voxel to compute the scores (and I do not know how we could vectorise this operation). However, this can be avoided when the F-contrast is a vector as the matrix at the right of the right-matrix division is a scalar. In this case, we can remove the for-loop over all voxels and use right-array division (./) to compute the scores. This strategy is currently used for computing the T-scores and was also used in older versions of swe_cp_wb for F-contrast of rank 1 (by computing the t-scores and then squaring them).

[nicholst]

I'm all for this, as good practice should be convincing more people to use two-sided tests, and hence F-tests for 1-DF contrasts.