Landmark and NBLAST sprint

[aschampion]

This is an issue for collecting discussion, small ease of use, and cleanup items for landmarks and neuron similarity given the recent push on these features.

Major

• [ ] Decouple transform creation from the actual transformation of skeleton sets (so transforms can be reused, and skeleton source UI can be reused for controlling transformed skeletons)
• [x] (assigned: @aschampion) Landmark transform MLS should support selection of affine (current), similarity, or rigid transformations
• [ ] Should investigate for same-volume transform when the groups have a mirror_of relation changing from the AABB-distance-piecewise-interpolation MLS affine, to using a single transform based on mutual mapping (i.e., for A <> B mapping, instead of an A->B and B->A, use AB->BA)
• [x] (assigned: @aschampion) When creating multiple mapping transforms, have a control that does name-based matching so that one can quickly checkmark select and add several mappings. This would take setting up transforms for, e.g., 0111-8 to 1018 from a ~100 click process to a < 10 click one.
  • [x] The name-based matching should have a regex and substitution string for both the source and target.
• [x] (assigned: @tomka) Bug when using transformed skeletons as the target set in a similarity query

Minor

• [x] (assigned: @tomka) Runtimes seem wrong in the neuron similarity widget (e.g., 443s for a run that took 3-4 hours)

[aschampion]

• [x] UI quibble: with filterable selects as they are, you can't see what's selected once the select collapses. This makes sense for compact uses like the 3D viewer volume select, but not for things like source/target group selection in the landmark widget.