Don't convert CIFTIs to int16

[tsalo]

Closes #977. The problem is that the CIFTI atlas files' contents are being converted to floats.

Changes proposed in this pull request

• Stop using DerivativesDataSink to write out atlas files. There is a bug in the base DerivativesDataSink where dlabel.nii CIFTIs with non-int16 data fail to convert to int16. Before, I was using nibabel to convert the datatype before feeding the file into the DerivativesDataSink. However, there appears to be a new bug in nibabel where that doesn't work, and changing the datatype in the header actually changes the data array's data type from normal ints to weird floats (e.g., 1 --> 1.46867437), which causes problems when we want to actually use those files.
  • Now, I have a function to compile the right filename and copy atlas file to the output folder.
• Also, replace the dseg.tsv file for the Gordon atlas, so the labels are more informative (i.e., match the names from Gordon333_FreesurferSubcortical.32k_fs_LR.dlabel.nii from https://sites.wustl.edu/petersenschlaggarlab/parcels-19cwpgu/). This also adds the community labels.
  • I couldn't find communities for Tian or HCP. The 4S atlases have community info for the Schaefer nodes.
• Replace the dseg.tsv file for the Glass atlas to add communities as well. Thanks to Jakob Seidlitz for the communities.

[tsalo]

I need to also add community affiliations to the Glasser dseg file, but there's an issue with the label files, so I'll circle back tomorrow.

[codecov]

Codecov Report

Attention: 1 lines in your changes are missing coverage. Please review.

Files	Coverage Δ
xcp_d/interfaces/connectivity.py	93.75% <100.00%> (+0.62%)	⬆️
xcp_d/utils/modified_data.py	93.75% <ø> (-0.99%)	⬇️
xcp_d/workflows/connectivity.py	96.93% <100.00%> (-0.07%)	⬇️
xcp_d/utils/atlas.py	98.14% <94.73%> (-1.86%)	⬇️

📢 Thoughts on this report? Let us know!.

[tsalo]

I just need to write a test for the new copy_atlas function.