Schaefer2018 7/17 Networks LUT files

[ppruc]

Adds LUT files to share/mrtrix3/labelconvert.

LUTs can be used to extract the relevant cortical grey matter parcellations from the Schaefer2018 atlas (https://github.com/ThomasYeoLab/CBIG/tree/master/stable_projects/brain_parcellation/Schaefer2018_LocalGlobal) and relevant subcortical grey matter parcellations from the default FreeSurfer segmentation (desikan_killiany).

As of 07/2023 the left accumbens area is incorrectly labeled as "Left-Accumbens-a" instead of "Left-Accumbens-area" in the official LUTs (see ThomasYeoLab/CBIG#49). Since labelconvert uses string recognition, I provided both strings ("Left-Accumbens-a"/"Left-Accumbens-area") and matched them with the corresponding label integer. The additional label can be omitted should the issue be resolved in the future.

[Lestropie]

1. First commit on the branch, 0fef112, only modifies test data submodule pointers, which is likely an unwanted regression. Rather than doing and then undoing this change, it would likely be preferable to start the changeset from scratch.

2. Looking around their repository a bit, I found this directory, which contains lookup tables with indices incrementing sequentially from 1. So it might be that the data proposed here is somewhat redundant.

   One difference between the two is that the files proposed here include not only cortical but also sub-cortical structures. This however assumes that one intends during connectome construction to use each discrete macroscopic sub-cortical grey matter nucleus as a single parcel. Particularly for higher resolution parcellations, this need not necessarily be the case; see for example the Melbourne Subcortical Atlas, which we used in the UK Biobank connectome project. For a hierarchical cortical parcellation such as Schaefer, it might make more sense to leave this up to the researcher. Long-term the goal is to allow researchers to specify their own lookup table that combines cortical and subcortical (and hypothetically multiple of such) parcellations, and feed multiple parcellation image - LUT input pairs to labelconvert, and have that command combine them appropriately; see ENH labelconvert: Multiple inputs #2481.

3. While there's been external contribution of lookup tables in the past, and the line counts involved have been fine, here because of the high resolution / hierarchical / multi-label nature of the data, the ratio of line count to manual labour investment might be a little too gratuitous. I would probably be more comfortable pulling some git gymnastics to attribute the bulk of the additions to @MRtrixBot but still a non-zero contribution from @ppruc.

   But as per point 2, I'm not sure it needs to be debated extensively as I don't know that the duplication of external data is justified in this case.

[Lestropie]

Closing in favour of #3043.