ENH: Enable sensor-specific OPM coregistration in mne coreg

[larsoner]

This PR does a few things:

• Adds a Kernel Flux MEG helmet based on their CAD
• Adds canonical sensor positions (based on the CAD)
• Adds a nonlinear matchied field displacement interpolator to take the CAD helmet to measured sensor positions. Assuming the number and placement of sensors EDIT: or other fixed points is fixed for a given system, this should allow generalization of a canonical/idealized CAD-based helmet to actual 3D printed or deformed helmets for individual participants.
• An option to show the MEG sensors as well: ENH: Add MEG sensor option to coreg #12098
• Disable backface culling on the MEG sensors if we haven't already: ENH: Add MEG sensor option to coreg #12098
• Add support for adjusting sensor alpha in mne.viz.plot_alignment by passing dict to meg
• Add some sample Kernel data with example

With this I get the following coreg:

Output

Getting helmet for system Kernel_Flux
    Deforming to match info using 39 matched points:
    1. Affine: 1.1°, 1.8 mm, 1.00× scale
    2. Nonlinear displacement: mean=2.7, max=8.5 mm

I think in follow-up PRs we'l want

1. A way to have mne coreg fix/adjust the head coordinate frame definition. Even with our existing OPM dataset from years ago, the head and MRI coordinate frames are taken as equal, which they shouldn't be. We should allow for using the current (in the GUI) head<->MRI transformation and current LPA/Nasion/RPA (from the MRI) to be used to adjust LPA/Nasion/RPA in info['dig'], which in turn should update the head coordinate frame definition, which in turn should change the rest of info['dig'] and info['chs][ii]['loc'] etc. We'll need some buttons and/or UI for this, but in principle I think this makes sense. We can test this easily with our existing OPM dataset by implementing it in the non-GUI Coreg object first.
2. A way to tell ICP fitting to use the MEG sensor locations as if they were digitization points
3. To add canonical helmets for other systems and/or adapt deformations to other schemes

Helps with #11276

[jasmainak]

@larsoner at Martinos we are currently at the stage of experimenting with subject-specific helmets ... I am not sure when/if we will transition to generic helmets. The question of optimal helmet needs some research ...

Whatever pipeline is developed in MNE should hopefully be a little flexible so that it's not too specific to a particular system. Things are still under flux in my opinion.

[larsoner]

Whatever pipeline is developed in MNE should hopefully be a little flexible so that it's not too specific to a particular system. Things are still under flux in my opinion.

Indeed, I think that's going to be the case with the plan above

[jasmainak]

Even with our existing OPM dataset, the head and MRI coordinate frames are taken as equal, which they shouldn't be.

The way I am thinking, one could define 5 or so index points on the helmet that need to be coregistered with the index points from the CAD file to get the dev_head_t ... this could be done using Procrusteas but potentially checked or adjusted using the GUI.

The head-MRI trans could be learned using the fiducials and digitized head surface points.

[drammock]

from the PR description

I think in follow-up PRs we'll want [...]

@larsoner can you open issues for items 1 and 2 (if you haven't already)? I think item 3 (add built-in models of other mfg's helmets) doesn't need an issue, we can just add them when folks ask for them.

[larsoner]

@drammock FYI I did end up adding a kernel dataset example. It should be building now!

[drammock]

reviewed just the new example this time.

[larsoner]

Think I got all the comments @drammock !

[larsoner]

@drammock took a stab at improving these, see if you think they make more sense now

[drammock]

that works, thanks!