Migrate pathfinding user guide off datashader

[brendancol]

Summary

• Migrate the pathfinding user guide notebook off datashader, part of removing datashader as a dependency (final packaging removal tracked in Drop datashader from packaging and install docs #3450).
• Build the line network from shapely LineStrings and rasterize it with grid.xrs.rasterize(...) instead of datashader.Canvas().line() / .points().
• Replace shade / stack / dynspread / set_background and the datashader colormaps with matplotlib (plot.imshow, ListedColormap, Patch legends).
• Switch the unweighted/weighted path colors from red/green to orange/blue so the figures read for colorblind viewers.
• Update the prose cell that named the old red/green colors.

The notebook runs top to bottom with no datashader import. The cost_distance and dask validation cells both still report Match: True.

Backend coverage

Docs-only change. No source code touched, so all four backends (numpy / cupy / dask+numpy / dask+cupy) are unaffected.

Test plan

• jupyter nbconvert --execute runs the notebook top to bottom with no errors
• No remaining datashader references in the notebook
• Replacement deps (geopandas, shapely, matplotlib) are present in the doc/tests extras the docs CI build installs

Closes #3460

https://claude.ai/code/session_01TEXdRkRqYJKvATjquz26Pa

[brendancol]

PR Review: Migrate pathfinding user guide off datashader

Docs-only change: one notebook, no source code touched. I read the full notebook from the PR worktree and re-ran it.

Blockers (must fix before merge)

None.

Suggestions (should fix, not blocking)

None.

Nits (optional improvements)

• The path overlays are single-pixel lines on the 600x400 terrain (cells 17 and 22). The old code used dynspread(..., max_px=2) to thicken them. The path is visible in the rendered figures but thin. If it reads too faint in the built docs, a small thickening or a thicker scatter would help. Not blocking.

What looks good

• The shapely segment list matches the original datashader xs/ys network exactly (checked the NaN-split segments against the original).
• grid.xrs.rasterize(lines, column="value", fill=0) is the right call: on-line cells get 1, everything else stays 0, which is what a_star_search(..., barriers=[0]) expects.
• Red/green path colors swapped for orange/blue, and the prose cell that named the colors was updated to match.
• import pandas as pd was dropped since the datashader DataFrame construction is gone and nothing else used it.
• The cost_distance and dask validation cells still print Match: True after the migration.
• datashader stays in the examples extra (other notebooks still use it); the packaging removal is left to #3450.

Checklist

• Notebook runs top to bottom with no datashader import
• Line network matches the original data
• Replacement deps (geopandas, shapely, matplotlib) available in the docs CI extras
• No orphaned imports
• Colorblind-safe path colors
• Backend parity / numerical tests: N/A (docs-only, no source change)
• README feature matrix: N/A (no new function)

[brendancol]

Follow-up review (after c9e6b68)

The one nit from the first pass is addressed.

• Added a path_overlay(path, grow=1) helper in cell 17 that grows the path footprint by one 4-connected pixel and returns 1.0 on the path / NaN elsewhere. Cells 17 and 22 now plot through it, so the terrain routes read clearly on the basemap.
• The helper is pure numpy (no new dependency) and preserves the path's dims and coords.
• Re-ran the notebook end to end: no datashader, and the cost_distance and dask validation cells still print Match: True.

No remaining blockers, suggestions, or nits.