Add Python vignettes for NNS R 13.0 curriculum

[OVVO-Financial]

This PR recreates the NNS vignette curriculum as Python-native documentation and executable examples.

The examples are based on the combined NNS vignette PDF and vendored R vignette sources, but are written against the public Python nns API. They assume the fresh R NNS 13.0 parity corrections from PR #3.

Validation:

• every example script runs
• docs example test passes
• R 13.0 smoke test passes
• invariants pass
• ruff, mypy, and build pass

Structure

docs/vignettes/         README + 00_overview .. 09_portfolio_stochastic_dominance
examples/vignettes/     one runnable script per vignette
tests/docs/             test_vignette_examples.py (runs each script, fails on nonzero exit)

Notes

• Translated from the vendored R vignette sources (tools/NNS/vignettes, tools/NNS/inst/doc), not OCR-copied; the ideas and examples are expressed idiomatically in Python.
• Deterministic where possible: seeded RNG, small data sizes for fast docs tests, numeric/parity-stable examples preferred over plotting.
• Stochastic routines (nns_meboot, nns_mc, the nns_ss CI, the nns_anova robust interval) are validated by structure and range, never by exact resampled values.
• The numeric ARMA, stack, and boost examples reuse the deterministic designs verified against live R NNS 13.0 in PR Fresh R NNS 13.0 parity regeneration and corrective Python fixes #3 (e.g. nonseasonal nonlinear ARMA [128.5, 113.5, 155.5, 213.6667]; stack reg/dim.red/stack; boost results/feature.weights/feature.frequency).
• No package runtime behavior was changed in this PR. Where a vignette idea touched an API surface, the examples were written to the real return shapes (verified against the installed package), not rewritten to hide gaps.
• No generated images or local artifacts are committed.

https://claude.ai/code/session_015pEPxpJWbiHizgcFEm1AWf

---

Generated by Claude Code