Skip mris_expand on pre-existing graymid/midthickness files

[effigies]

With #476, I'm remembering that mris_expand gets called every time the scratch directory gets wiped out, even though the output (midthickness files) exists. This is a roughly ~12min step. I'm not sure what the best way to skip this would be.

One option would be: Add a graymid/midthickness output to nipype.interfaces.io.FreeSurferSource. Then subclass MRIsExpand to take that field as an input. If it exists, change the command to cp $GRAYMID $HEMI.midthickness.

In general I don't love that we're writing into a subject directory, but for reasons explored in #398 (comment) it's a least-bad option. But especially if somebody provides a pre-computed midthickness file, I think it's better not to replace it, and the above proposal overwrites it, but with a copy of itself, so it's pretty tame.

Any other thoughts?

[effigies]

I don't know if it's a processor situation or what, but mris_expand takes a bizarre amount of time on Circle.

From a recent log:

170425-12:47:17,808 workflow INFO:

	 Executing node _midthickness1 in dir: /scratch/fmriprep_wf/single_subject_01_wf/anat_preproc_wf/surface_recon_wf/midthickness/mapflow/_midthickness1

...

170425-13:14:10,48 workflow INFO:

	 [Job finished] jobname: _midthickness1 jobid: 155

That's nearly a half hour, the last fifteen minutes spent waiting because the rest of the workflow (except dependent jobs) are finished.

[effigies]

First step: nipy/nipype#1972