Bundled OpenAI Docs skill has stale GPT-5.6 Luna and Pro mode facts

Codex rust-v0.145.0 bundles three defects in the OpenAI Docs skill:

1. codex-rs/skills/src/assets/samples/openai-docs/references/upgrading-to-gpt-5p6-sol.md says GPT-5.6 Luna has a 400K context window and omits Luna from the >272K long-context surcharge. The current model page says 1,050,000 context, 922,000 max input, and the same >272K pricing rule: https://developers.openai.com/api/docs/models/gpt-5.6-luna
2. The same guide says Pro efforts begin at medium. Current migration guidance says reasoning.mode and reasoning.effort are independent; GPT-5.6 supports none, low, medium, high, xhigh, and max, with medium only the omitted default: https://developers.openai.com/api/docs/guides/latest-model#update-api-and-model-parameters
3. codex-rs/skills/src/assets/samples/openai-docs/scripts/resolve-latest-model-info uses set -u but expands $HOME unconditionally. With HOME and NODE unset and no node on PATH, it exits with HOME: unbound variable before trying valid /opt/homebrew/bin/node, /usr/local/bin/node, or /usr/bin/node fallbacks. A direct reproduction is: env -u HOME -u NODE PATH=/usr/bin:/bin sh resolve-latest-model-info --source references/latest-model.md on a machine with /opt/homebrew/bin/node.

These claims and the wrapper matter because they are the fallback when live docs retrieval or the normal Node path is unavailable. Please update the repeated Luna context/pricing references, remove the Pro-mode minimum-effort restriction, and guard the home-relative runtime candidates.

Observed in stable rust-v0.145.0; the same source is present on main.