refactor: merge octomap examples into a camera-synced 2x2 viewer

[wkentaro]

Summary

• Replace insertPointCloud.py and insertPointCloudColor.py with a single examples/pointcloud_to_octree.py that builds an OcTree and a ColorOcTree from one scan and renders pointcloud / occupied / occupied (color) / empty in a 2x2 window.
• The four panes share one trackball, so dragging, zooming, or panning any pane moves all four cameras together.
• Extract the rendering machinery (the synced widget and the grid window) into examples/viewer.py, leaving the example as a concise walk through the octomap API.

Test plan

• ruff check examples/ and ruff format --check examples/
• Drove the example headlessly: a drag on one pane moves the camera and all four panes sync to the same pose (single shared trackball), confirmed by a rendered screenshot of the 2x2 grid.

[wkentaro]

This was generated by AI during PR processing.

Verdict: `recommend-merge`

Own PR, full-finalize pipeline. main gained 0 commits since the merge-base, so no rebase was needed and the existing green CI (9/9 checks) reflects the merged result; no push happened this tick.

Review (report-only /code-review, no meaningful findings):

• The _extract_colored_voxels index rewrite (np.indices(...).reshape(3,-1).T) is provably equivalent to the old np.column_stack(np.nonzero(np.ones(...))) (verified numerically).
• SyncedSceneWidget is sound: trimesh SceneWidget.__init__ calls reset_view(), so view["ball"] exists before sync(), and reset_view is not called on resize, so the shared trackball persists for the window lifetime.
• The OcTree (nonnan) vs ColorOcTree (z < 2) mask split is intentional and documented inline.

/recommit: the two commits (refactor(examples) then docs) are already a clean, logical sequence; no reshaping needed. /verify: skipped (GUI viewer, not headlessly smoke-testable).

No follow-up or idea issues warranted.