Refactor CIFTI bold -> label subcortical resampling

[mgxd]

I believe here's where you sample the subcortical structures from the BOLD data. Now that we have learned so much from nitransforms, you can do this step without the resampling above.

The pseudocode would be:

ijk_labels = np.where(label_data > 0)  # I don't remember if you need to set an upper bound threshold
ijk_bold = np.linalg.inv(bold_img.affine).dot(label_img.affine.dot(ijk_labels))  # You'll need to take care of ijk_labels, add a ones column and probably transpose here.
bold_values = scipy.ndimage.map_coordinates(bold_data, ijk_bold, order=3)

Originally posted by @oesteban in #417