Update extract.ai recipe notebook after PY 1.20.5 release #1231

Outcome

After the coordinated PY 1.20.5 release, update docs/examples/extract_ai/extract_ai_recipe_examples.ipynb so its runnable YAML examples match the released extraction behavior.

Eric explicitly deferred this cleanup until after the release on October 11, 2026. This issue is not a PY 1.20.5 release blocker. Runtime Luna defaults and flat/nested reasoning support are already implemented through #1200 and #1226.

Release planning: #1227. Original Luna migration assessment: #1225.

Prepared work and scope

A local documentation-only checkpoint already exists: ed69a56906a0d72b46798e472be9133264f29132 on codex/model-defaults-luna. It changes only the notebook above. Treat it as a starting point to review against current main, not as a fully completed or live-validated notebook.

• Update outdated model examples and model comparisons to Luna/Sol where appropriate.
• Demonstrate omitted reasoning, flat reasoning: none, equivalent nested reasoning: {effort: none}, and explicit effort overrides.
• Remove the obsolete explanation that unsupported reasoning is silently dropped; preserve the released behavior where provider rejection remains visible.
• Use the current protocol options for Responses and legacy Chat Completions examples.
• Remove stale captured warnings, model outputs, and benchmark timings from changed examples. Do not present old results as measurements of the new models.
• Review the whole notebook for other stale model/parameter references, including cells not changed by the checkpoint.

Acceptance checklist

• Begin after the PY 1.20.5 release; create a focused branch from freshly fetched main and recover only the wanted notebook changes.
• Confirm the examples and explanations against current WranglesPY implementation/schema, preserving both reasoning YAML forms and explicit overrides.
• Validate notebook JSON, changed code cells, and embedded YAML without credentials or embedded secrets.
• Either regenerate affected outputs through explicitly reported live runs or leave them cleared; distinguish local/static validation from live provider results.
• Open a documentation-only PR to main with the changed notebook and an accurate validation summary.

Do not reopen the broader configuration/compiler redesign, modify production defaults, migrate saved customer definitions, or include deferred generate.ai work as part of this cleanup.