Feature request: non-rigid transforms

It would be incredibly powerful if neuroglancer supported simple non-rigid transforms like thin plate spline transforms from user-provided landmarks (via CSV/URL).

Possible applications:
- co-visualizing neurons from another dataset/space (e.g. pulling hemibrain neurons into FlyWire)
- mirroring of neurons along symmetry axes (e.g. for left/right matching)

I'm unfortunately not clever enough to implement this myself but I thought I'd open this issue to put the idea out there.