Minimum image regression relies on distinction between "accepted" and "ignored" components

[tsalo]

Summary

I'm working on refactoring the global signal regression functions in order to better understand them (particularly for #1071), and I noticed that gscontrol.minimum_image_regression specifically uses "accepted" components and not "ignored" components. Since we eliminated the "ignored" classification in #756, MIR likely doesn't work the way it was intended anymore.

We could probably leverage the classification tags, but those vary by decision tree and are essentially up to the user.

[tsalo]

Ah, I see that "low variance" and "accept borderline" only exist as classification tags in the meica and tedana_orig trees.