ENH: Speed up and clean up 3D code

[larsoner]

While I was digging through mne-tools/mne-kit-gui#68, I realized our codebase uses a lot of older PyVista idioms that we can clean up. But that in turn inspired some other changes:

1. Use PyVista 0.43+ APIs (like Lights) and remove cruft related to older versions, plus params/modes we never actually use (resolution, polygon_offset)
2. Single-sensor plotting (e.g., with 300+ sensors) was a huge slowdown -- so now we group sensors by color and concat into a single mesh (very large plot_alignment speedup)
3. Switched to MNE-based mesh decimation (~5s to plot a brain with a silhouette to ~1.4s) by default that makes use of the existing/known FreeSurfer mesh decimation
4. Fixed a sneaky issue with foci removal that was causing cleanup errors depending on test run order (sometimes!)
5. Streamlined the _assert_no_instances by avoiding multiple gc.collect() calls (by using request), and by keeping track of Brain instances directly when in testing mode (which greatly reduces the search surface when
6. Moved assert_no_instances code out of MNE and into https://github.com/mne-tools/refleak and added them as doc-build and test deps

I used Claude Sonnet 5 to help with this (especially the profiling), but I understand and vouch for all code.

[wmvanvliet]

At some point, it may be nice to split of all the referrer tracking code into its own package. This looks very useful also for debugging other projects with memory leaks.

[wmvanvliet]

I did some simple testing and plotting sensors is now much faster. Much appreciated. I just need to find a new way now to color selected sensors in the dipole fitting GUI.

[larsoner]

I just need to find a new way now to color selected sensors in the dipole fitting GUI.

Okay when coding this I wasn't 100% sure if this was needed. But it does provide a motivation for implemeting the vtkGlyphActor3D method -- that should make it much simpler to color individual actors. So let's not merge for now, I'll fix that here!

[larsoner]

At some point, it may be nice to split of all the referrer tracking code into its own package. This looks very useful also for debugging other projects with memory leaks.

Yeah pyvista and pyvistaqt use very similar code (I think I adapted it from MNE originally). I'll probably tackle that once this lands

[larsoner]

I'll check the outputs soon and then merge assuming everything still works. @wmvanvliet the vtkGlypthActor3D has been implemented, and there's code in there that shows you can update the colors easily now by updating the point_scalars["colors"] array, and marking the object as modified!

[larsoner]

@wmvanvliet I've pushed this to https://github.com/mne-tools/refleak and https://pypi.org/project/refleak/#history and started using it here. I'll iterate until it works but hopefully it should. I'll make corresponding PRs to pyvistaqt and pyvista soon!