Add get_montage support for fNIRS

[rob-luke]

Reference issue

Follow up to #10555
Closes #10585

What does this implement/fix?

Fixes get_montage for fNIRS data so that a user can run raw.set_montage(raw.get_montage()).
fNIRS data has multiple channels between each optode (source, detector). This was already handled with set_montage in #9141

[rob-luke]

With this branch I can run the following script....

import mne
import os.path as op

testing_path = mne.datasets.testing.data_path(download=False)
lumo_fname = op.join(testing_path, 'SNIRF', 'GowerLabs', 'lumomat-1-1-0.snirf')

subjects_dir = op.join(mne.datasets.sample.data_path(), 'subjects')
raw = mne.io.read_raw_snirf(lumo_fname, preload=True)

coreg = mne.coreg.Coregistration(raw.info, "fsaverage", subjects_dir, fiducials="estimated")
coreg.fit_fiducials(lpa_weight=1., nasion_weight=1., rpa_weight=1.)

mtg = raw.get_montage()
mtg.apply_trans(coreg.trans)
raw.set_montage(mtg)

brain = mne.viz.Brain('fsaverage', subjects_dir=subjects_dir, background='w', cortex='0.5', alpha=0.3)
brain.add_sensors(raw.info, trans="fsaverage", fnirs=['sources', 'detectors'])

[rob-luke]

@agramfort and @larsoner could you please review.
This achieves my goal, but I am not sure if this was the best way to implement this. Feedback is appreciated.

[rob-luke]

FYI: @samuelpowell

[agramfort]

just a few nitpicks to better understand. 🙏 @rob-luke

[rob-luke]

Thanks for the feedback @agramfort. I have tried to explain my reasoning for the approach I have implemented. But if you see a cleaner way we can implement this that is more aligned with the other modalities then I am eager to try it.

[larsoner]

I pushed a tiny commit to skip more tests on Windows to help avoid timeouts. Other than the remaining multimodal issue, LGTM!

[agramfort]

@larsoner I let you merge if happy

[larsoner]

Just pushed a commit to hit the raise path, will merge once green -- thanks in advance @rob-luke !