Add visualization for cortical depth profiles

[sunaguo]

Visualize laminar data with quickflat tool or webgl.
Closes #731.

quickflat/laminar.py

Given two points (flatmap coordinates) on the surface, grab cortical depth data along the line that connects the two points.

webgl

Ports quickflat/laminar.py into the browser: draw a line across the flatmap and get the W x H matrix spanning that line (gamma) and cortical depth (alpha, pial to white matter), sampled from the volume.

Everything is computed client-side, nothing is round-tripped to python:

laminar.js FlatIndex is a uniform grid over the flatmap triangles,
standing in for the scipy Delaunay/trifinder (the flatmap
already carries its own triangulation). For each of the W+1
columns it locates the flatmap point and barycentrically
interpolates the pial position, the white matter position
and the two surface areas -- _column_geometry(), once per
column rather than once per pixel.

Shaders.laminar Per fragment, turns alpha into an equivolume depth
(_depth_blend), mixes pial/wm to get that pixel's own
anatomical coordinate, pushes it through volxfm and samples

• colormaps it via the same colorlut() the surface shader
  uses. So each panel pixel maps to its own point in the
  volume, and vmin/vmax/colormap/sampler follow the dataview.

The line lives in the shared, nudged flatmap space (the uv attribute), which is the same merged space laminar.py triangulates. That makes the profile independent of the view state and lets a line run across both hemispheres, with columns landing in the gap coming out transparent.

The line is drawn into a flatmap-space texture that the surface shader composites through vUv alongside the ROI overlay, so it sits on the cortex in every view -- flattened, inflated or folded -- with the right occlusion. Endpoints are dragged from a screen-space overlay that is only active while the brain is flat.

UI: a "depth profile" folder (show / equivolume / samples / depths / reset_line, shortcut J) and a draggable panel. The readout reports the endpoints in raw flatmap units, i.e. the arguments make_laminar_profile() takes.

Verified against cortex.quickflat.laminar on S1: the columns computed in the browser match _column_geometry to within CTM quantization (<0.002 voxels). The equivolume spacing uses the viewer's smoothed per-vertex areas rather than laminar.py's raw triangle areas, so the profile's depths agree with what the surface shader draws at the same depth.

[mvdoc]

Something funky is happening with roi labels with this PR