Ship complete teaching notebooks with worked "Example solution" cells (#6)

[jonfroehlich]

Closes #6.

Per issue #6, the published notebooks are now complete teaching artifacts: each keeps its exercise prompt / "your turn" framing, but adds a clearly-labeled ### Example solution (markdown + worked, executed code) immediately below. New graded assignments will be authored separately.

What changed

• Tutorials (3 signals notebooks): model answers for the make-your-own-chord, cross-correlation-subsequence, and quantization exercises.
• StepTracker: consolidated the Exercises + WithExampleSolution pair into a single notebook (deleted the redundant solution file). Folded in the offline detector and authored a genuine real-time solution (buffered, ongoing detrend + smoothing) — the old solution file had left the real-time TODO unsolved.
• GestureRecognizer-ShapeBased: implemented generate_kfolds from scratch (the 2020 "private" copy was also an empty stub), added a DTW-based matcher (~91% vs ~60–70% for raw Euclidean), and a fuller preprocessing demo.
• GestureRecognizer-FeatureBased / Feature Selection: example answers for the frequency-domain and 2D feature "your turn" cells plus a detrend + smooth preprocessing demo. (Cross-user classification was already implemented in v2, so it's left as-is.)
• CLAUDE.md: documents the new "complete artifact" convention and the StepTracker consolidation.

Notes for reviewers

• All edited notebooks were re-executed with saved outputs (the textbook renders committed outputs), so diffs are large — mostly regenerated output, not hand edits.
• The 2020 Dropbox "private" notebooks were not a clean answer key (generate_kfolds was an empty stub there too), so several answers are authored fresh rather than copied.

Verification

• pytest tests/ → 29 passed
• pytest --nbmake fast lane (Tutorials + StepTracker) → 8 passed
• pytest --nbmake GestureRecognizer lane → 3 passed

🤖 Generated with Claude Code