Overhaul of the carpet plots

[oesteban]

We need several improvements to be done to the carpet plots:

• Improve the readability (and perhaps a colorblind-safe color scale) of the indication of segments (WM, cortical GM, CSF, etc.), by making the indicator thicker, perhaps adding a minimal separation between segments, etc. Alternatively, add textual legends
• Better labeling of axes
• Ward clustering
• Reconsider the standardization strategies (perhaps all data standardized the same way, rather than region-wise?)
• Ensure we are dropping nonsteady states (I think we just merged that with ENH: carpet plot tweaks #617).
• Tweaks to the timeseries represented on top:
  • Global signals, FD+DVARS, CompCor(s) can be plotted sharing x-axis.
  • Perhaps plot a kernel density estimation on the right showing the distribution of confound timeseries.
• Chris' idea of delineating CompCor ROIs on the carpet.

[effigies]

For colorblindness, I would default to Seaborn's colorblind palette: https://seaborn.pydata.org/tutorial/color_palettes.html#qualitative-color-palettes

And here's a gist with simulated contrasts under different colorblindness conditions: https://gist.github.com/mwaskom/b35f6ebc2d4b340b4f64a4e28e778486

I believe we need 7 colors if we move to CIFTI and splitting the aCompCor mask out of WM+CSF: crown, lh, rh, subcortical, cerebellum, WM+CSF-aCompCor, aCompCor