Add voxel-wise option to Brain_Data.scale

[ejolly]

Looking back at how the "big 3" do scaling, it looks like both FSL and SPM do what we do, scaling the full time-series by a single value: the grand-mean across the acquisition. AFNI, on the other hand, has some compelling reasons to instead, normalize each voxel separately by its own acquisition mean. In practice, the difference is likely not that huge but could be easily added with an optional axis flag. If we do add this, we need to account for voxels who's mean maybe 0 (e.g. out of brain voxels) as scaling will produce NaNs

b = Brain_Data('some_file.nii.gz')
b.scale() # default axis=None, scale each voxel's time-series by the grand-mean
b.scale(axis=0) # AFNI style, scale each voxel by it's own mean