WIP: Add Ipyvolume 3d backend

[GuillaumeFavelier]

What does this implement/fix?

The goal of this PR is to add a Ipyvolume-based 3d backend to the visualization capabilitites of MNE and inside a jupyter notebook in particular.

Additional information

It is a work in progress, some features are still missing or incomplete as of now.

[larsoner]

1. This:

   t = mne.viz.plot_alignment(info, trans, subject=subject, dig=True,
                             meg=['helmet', 'sensors'], subjects_dir=subjects_dir,
                             surfaces=['head', 'inner_skull', 'outer_skull'])
   

   Only shows one surface:
   
   Perhaps depth testing needs to be disabled?
2. The surface above looks weird. Perhaps alpha blending and/or backface culling needs to be enabled?
3. The surface is not smooth shaded, it's flat shaded. If this is still WIP at the ipyvolume/pythreejs end, then put this in the Notes section as suggested above about **Backend-specific notes**.
4. Adding a source space:

   src = subjects_dir + '/sample/bem/sample-oct-6-src.fif'
   t = mne.viz.plot_alignment(info, trans, subject=subject, dig=True,
                             meg=['helmet', 'sensors'], subjects_dir=subjects_dir,
                             surfaces=['head', 'inner_skull', 'outer_skull'], src=src)
   

   Does not plot anything, and does not emit an error in Python, but the Chrome console says:
   

[GuillaumeFavelier]

Thanks for the suggesting 1. and 2., I'll try those flags out right now.
3. is normally already written in the 3d backends overview in the section 'Smooth Shading' normally.
Thanks for reporting 4. I'll investigate what's going on here too

[lgtm-com]

This pull request introduces 1 alert when merging 3ce92e0 into c2b9176 - view on LGTM.com

new alerts:

• 1 for Wrong number of arguments in a call

[GuillaumeFavelier]

Thanks for the refactoring @massich !

The recent modifications should fix the scenario 4. you had @larsoner (the quiver3d() function with the cylinder glyph was not supported yet).
Now solving 1. and 2. is a matter of finding a good set of parameters to obtain decent transparency. I'm working on it.

[lgtm-com]

This pull request introduces 4 alerts when merging 442f8b3 into f4b66b5 - view on LGTM.com

new alerts:

• 3 for Unused local variable
• 1 for Unreachable code

[larsoner]

@GuillaumeFavelier the rendering does not need to be perfect, just need to be able to see the three surfaces. Over in IPyvolume you show that even though it's not 100% correct you can see three overlaid meshes. Is that the case here now, too?

[GuillaumeFavelier]

With the latest version, even with backface_culling and my tests on depthTest and depthWrite I get something close to this (we still can't see the brain surface inside without zooming):

The most satifying resut I got was inverting the processing order during depth test but this is a workaround that works only for this example:

My next test is about the rendering order which could be another promising way to achieve a decent result.

[larsoner]

@GuillaumeFavelier what do you think we should do here? Abandon for now due to alpha limitations?

[agramfort]

yes leave it for later for now

[larsoner]

Okay I'll close to clear up a bit of backlog