Tutorial plotting API

This package turns the versioned JSON/CSV artifacts emitted by the native Rust application tutorials into consistent Matplotlib figures. It does not put Python callbacks in the optimization hot path.

From public/tutorials/python:

python -m venv .venv
.venv/bin/python -m pip install -r requirements-lock.txt
.venv/bin/python -m pip install --no-deps -e .
.venv/bin/python -m pytest
.venv/bin/python render_all.py --check
.venv/bin/python check_docs.py

Render one run:

.venv/bin/fcmaes-tutorial-plots \
  ../rapier-trebuchet/results/quick/qd/run.json \
  --output-dir ../rapier-trebuchet/images/quick-qd

The API also accepts result arrays from the optional PyO3 bindings through pareto_from_arrays and qd_from_archive.

render_all.py --write regenerates every figure from discovered schema-v1 manifests and invokes any tutorial-specific plot_results.py --write script. The room-ventilation and HPO tutorials use that extension because their publication figures combine evidence that is broader than one optimizer manifest. A tutorial whose artifact axes are incompatible with the common renderer may also add a .custom-renderer marker. Its manifests are then skipped by the common renderer while its deterministic plot_results.py remains part of the same write/check pass. --check renders into temporary directories and compares bytes, so volatile Matplotlib metadata cannot create silent documentation drift. requirements-lock.txt fixes the complete plotting stack used for regeneration and byte-for-byte validation; use it whenever checked-in figures are updated. check_docs.py verifies that every local Markdown link and image target exists. When public/ is the Git repository root, as it is after publication and in CI, the check also requires every target to be present in the Git index. This prevents locally generated but uncommitted result files from masking broken README evidence links.