fix(models): correct GPT-5.6 context windows

[kidclone3]

Summary

• correct GPT-5.6 Sol, Terra, and Luna bootstrap context windows from 372,000 to 272,000
• update bootstrap catalog contract tests

Evidence

Authenticated live /codex/models catalog fetches from three Plus accounts and one Pro account all reported 272000 for gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna. The local spike artifacts containing that evidence are intentionally untracked and are not part of this PR.

Validation

• uv run pytest tests/unit/test_model_registry.py::test_bootstrap_models_include_representative_upstream_metadata
• uv run pytest tests/integration/test_v1_models.py::test_backend_codex_models_uses_bootstrap_upstream_metadata
• uv run ruff check app/core/openai/model_registry.py tests/integration/test_v1_models.py tests/unit/test_model_registry.py
• uv run ruff format --check app/core/openai/model_registry.py tests/integration/test_v1_models.py tests/unit/test_model_registry.py

[Soju06]

Thanks — the value itself checks out: upstream corrected GPT-5.6 Sol/Terra/Luna to 272000 in codex-rs/models-manager/models.json starting at rust-v0.145.0 (372000 was only ever in rust-v0.144.1, which is what our catalog pinned). Three things before this can merge:

1. OpenSpec (hard gate): openspec/specs/model-catalog-compat/spec.md normatively pins context_window/max_context_window of 372000 mirroring rust-v0.144.1. Please add an OpenSpec change folder (e.g. openspec/changes/fix-gpt56-context-window/ with proposal + delta spec) that re-pins the requirement to rust-v0.145.0 (or later) with 272000, so spec and code stay in sync.
2. Contributors attribution CI: you're not in .all-contributorsrc yet — please add yourself (all-contributors) so the check passes.
3. Branch update: the Beta release guard failure is a known false positive on branches behind the latest release commit — please rebase/update onto current main.

Also, in the test comments, prefer citing codex-rs/models-manager/models.json at rust-v0.145.0+ over untracked live-fetch artifacts — it's reproducible for reviewers. I'll kick off @codex review once the branch is updated.

[chatgpt-codex-connector]

To use Codex here, create an environment for this repo.

[kidclone3]

Addressed the merge blockers.

• Added openspec/changes/fix-gpt56-context-window/, re-pinning the normative GPT-5.6 bootstrap source to codex-rs/models-manager/models.json at rust-v0.145.0 and requiring context_window / max_context_window = 272000.
• Replaced the untracked live-fetch test comments with the reproducible rust-v0.145.0 source reference.
• Added @kidclone3 to all-contributors; the coverage check passes.
• Updated the branch with current main without rewriting remote history; it is now 0 commits behind.

Validated: focused GPT-5.6 bootstrap tests (2 passed), ruff format --check, ruff check, strict change validation, full spec validation (49 passed), and all-contributors coverage.

[Komzpa]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f43fefb80c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Komzpa]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c04455d672

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".