Determine how best to divide 'minimal', 'resampling', and 'full' levels

[tsalo]

Summary

#338 introduces the --level parameter, which matches fMRIPrep's new approach.

minimal:

• transforms
• aslref files
• brain mask file
• native-space CBF

resampling:

• native-space preprocessed ASL
• native-space CBF

full:

• not native-space CBF, unless requested
• standard-space outputs
• atlases
• parcellated outputs

What about confounds and CBF QC?
Also, CBF isn't a simple transform of ASL, so should I include that in minimal?
Finally, the ASL time series single-shot resampling is really computationally intensive, so maybe I should drop it from anything but full? Or maybe add a parameter that controls whether it's written out or not?

[tsalo]

I think I need to write out aslref-space CBF maps when the workflow is set to minimal or resampling.