test-containers-lectures: absorb transient raw-GitHub 429s when building lecture repos

A test-containers-lectures.yml run failed on lectures/hansen_singleton_1983.md in QuantEcon/lecture-python.myst — CellExecutionError / HTTP Error 429: Too Many Requests: a hidden cell fetches its dataset from raw GitHub at build time (lectures/hansen_singleton_1983.md:1432-1439).

The matrix builds several repos at once from GitHub's shared runner IPs, so raw GitHub rate-limits it — non-deterministically: the same notebook passed on the other container in the same run. It surfaced during the Anaconda 2026.06 migration (#95), but the cause is the fetch, not the migration or pandas.

Lecture-side fixes do not work: local _static/ paths break Colab, since QuantEcon/lecture-python.notebooks is a flat .ipynb directory with no _static/ (QuantEcon/lecture-python.myst#972, closed for that reason). It is not lecture-specific either — lectures/hansen_singleton_1982.md:1000, lectures/pandas_panel.md:68,188,505 and lectures/phillips_drifts_volatilities.md:103 fetch over raw GitHub too. Moving datasets to data-lectures does not help: pandas_panel.md already reads from that repo over the same github.com/.../raw/ host.

No retry, backoff or 429 handling exists here. The only mitigation, per-repo serialization at .github/workflows/test-containers-lectures.yml:46-48 (v0.7.0, re-scoped by run id in v0.9.0, #103), pairs the two containers for one repo; the three repos at :55-60 still build in parallel. Flakes also file tracker noise: on any non-workflow_dispatch failure (:138), notify-failure (:132-154) runs scripts/create-ci-failure-issue.sh → "🔴 CI failure: Test Container Builds".

Fix: retry the HTML build stage. With execute_notebooks: "cache" (lectures/_config.yml:21) and lecture-repo/_build/.jupyter_cache persisting across steps in the job — why stages 2 and 3 already skip execution — a retry re-executes only the failed notebooks, not the full 30-60 minute run, so no need to gate on the 429 string.

• Give the "Build HTML (Execute Notebooks)" step (:95-102) an id and continue-on-error: true, then a conditional second attempt on steps.<id>.outcome == 'failure'
• Give the retry a distinct failure-artifact-name — build-lectures uploads execution reports under that name on failure (build-lectures/action.yml:38,179), so a repeated name collides with the first attempt's artifact
• Validate with a maintainer workflow_dispatch after merge — it triggers only on workflow_run and workflow_dispatch (:22-27), so a PR cannot exercise it

Acceptance: a single transient 429 no longer fails the matrix; a persistent failure still does.

While in this workflow: restore the missing matrix repo

lecture-python-programming was dropped from the matrix (:54-64) in an unannounced swap in 81e96ff — unlike lecture-jax, which is commented out at :61-64 with a TODO and an issue link. containers/VALIDATION.md:25,42,46,50 and TESTING.md:11 still advertise the 4-repo matrix, so the only place lecture content executes on the images silently lost a quarter of its coverage while the docs kept claiming it.

It belongs in this PR rather than its own: it is four lines in the file you are already editing, and the repo has six build-time raw-GitHub fetches plus yfinance calls, so the retry above is the prerequisite for adding it without immediately reintroducing the flake. Note it is also the repo pinning anaconda=2026.07 against the containers' 2026.06.

• Add the lecture-python-programming matrix row (current repo name, no .myst suffix)
• Correct the 4-repo claims in containers/VALIDATION.md and TESTING.md to match

Start with .github/workflows/test-containers-lectures.yml, especially the Build HTML (Execute Notebooks) step and matrix rows, then inspect build-lectures/action.yml for artifact naming. Update the workflow to retry failed HTML builds and restore the lecture-python-programming row, then align the four-repo statements in containers/VALIDATION.md and TESTING.md. Done means transient 429s are absorbed while persistent failures still fail and the matrix documentation matches.