Add NanoVDB TEACHME interactive tutorial + docs-compile test harness

[harrism]

Adds doc/nanovdb/TEACHME/ — an interactive, LLM-guided tutorial for the
NanoVDB user API, plus a test harness that keeps it from drifting from the
real headers.

• User lesson (nanovdb_user_lesson.md) — 10 modules: sparse-grid
  foundations, GridHandle/IO, ReadAccessor, NodeManager, math + sampling +
  HDDA, GPU kernels, GPU grid builders (PointsToGrid/MeshToGrid), topology
  operators + data re-homing (Injection), IndexGrid/VoxelBlockManager,
  and OpenVDB ↔ NanoVDB conversion.
• Cheat sheet (nanovdb_user_cheatsheet.md) — one-page API reference.
• Capstone (capstone/) — a GPU level-set ray-march renderer (scaffold +
  reference + CMake), rendering an SDF sphere on the device.

Why a test harness

Markdown code in docs is never compiled, so it silently rots as the API
changes. test/ extracts every C++/CUDA block and compiles it (compile-only)
against the real NanoVDB headers, and validates every #include / prose header
reference. ~64 blocks compile green; run test/run.sh. This is in the same
spirit as the doc-quality CI gates being added in #2217.

A CI workflow (.github/workflows/teachme.yml) runs the harness whenever the
NanoVDB headers (nanovdb/nanovdb/**) or the tutorial change, so an API change
that breaks a lesson is caught automatically. It's compile-only (~90s on a
standard runner, no GPU) and reuses the existing aswf/ci-openvdb CUDA image.

Notes

• Docs + harness + a CI workflow that runs it — no changes to NanoVDB headers.
• The harness is compile-only by design (catches API drift); the capstone is
  the runnable end-to-end check.
• Authored interactively with Claude (see commit co-author trailer).

[kmuseth]

This looks great - nice work! Please update pendingchanges/nanovdb.txt

A few high-level questions: it's unclear exactly what and how this documentation is intended for. More to the point, is it for novice developers or AI agents? I was expecting to see this in a top-level README, but I couldn't find it.

[harrism]

Thanks Ken — pendingchanges added in 94c0469, as a New Features bullet in pendingchanges/nanovdb.txt.

On the audience question: it's for human developers who are new to NanoVDB, not for AI agents. The LLM is the delivery mechanism, not the reader. The lesson file is written as an instructor prompt — the agent reads it and then teaches you the API interactively, module by module, quizzing as it goes and adapting to your answers. You are the student; the agent is the teacher. It is explicitly not agent-facing reference material like a CLAUDE.md or an MCP schema.

The reason for that shape is that the agent can read both the lesson and the NanoVDB headers in the same session, so it can verify API details against the real source and debug your exercises live — which a static tutorial can't do, and which is also why the tutorial can't drift silently (the harness compiles every code block against the real headers).

That framing is in doc/nanovdb/TEACHME/README.md, but you're right that it isn't discoverable — you had to find the directory first. The PR currently links it from nanovdb/nanovdb/Readme.md and doc/nanovdb/doc.md, which are the two NanoVDB doc entry points; the repo-root README.md has no per-component documentation index, so there was no obvious slot there.

Happy to fix the discoverability whichever way you prefer — tell me which you meant by "top-level README" and I'll do it:

• lead doc/nanovdb/TEACHME/README.md with an explicit "who this is for" paragraph;
• promote the summary into nanovdb/nanovdb/Readme.md rather than a one-line bullet;
• add a documentation section to the repo-root README.md and list it there.