Reinstate learning_approximation — untranslated, dropped from _toc.yml, four cross-references neutralised

[mmcky]

Carried in QuantEcon/project-translation NEXT-STEPS since 2026-07-27 without a tracker; filing so it survives the planning-file prune.

The lecture failed structural parity deterministically on two consecutive syncs (QuantEcon/lecture-python.myst#1004, QuantEcon/lecture-python.myst#1007 — the QuantEcon/action-translation#118 fence-wrap signature) and was dropped from _toc.yml during the 2026-07-27 hand repair of #207, with its four {doc} references neutralised. The edition ships without it; the English source carries it at ~31 KB.

Reinstatement is blocked on QuantEcon/action-translation#118 (or a hand-seeded init -f with manual wrapper repair, as was done for the tenth Phillips lecture). When it lands: restore the _toc.yml entry, re-activate the four references, and verify with the strict -n -W build. Until then this lecture is also the cheapest deterministic reproducer the estate has for #118 — worth using as the validation case when that fix is attempted.

[mmcky]

Duplicate of #210, which predates this by nine days and carries the blocking measurement (5/15 structural pass rate with the localisation rules against 14/15 without, Fisher exact p = 0.002). This issue's unique content — the myst#1004/#1007 deterministic-failure links and the use-as-validation-case note for action-translation#118 — is folded into #210 in #210 (comment).