fix(core): honor Codex input limits

[magoz]

Issue for this PR

Closes #38851

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Preserves models.dev's separate input limit through LLM routes. For ChatGPT OAuth, it overlays canonical openai-codex limits onto matching OpenAI models; API-key limits stay unchanged. Compaction uses input - buffer, and the TUI reports usage against that effective input window.

This makes GPT-5.6 Sol compact at 252K (272K input minus the existing 20K buffer), before the Codex backend ceiling. Depends on anomalyco/models.dev#3754.

Supersedes conflicting #38987 with the same Kit-authored patch rebased onto current v2.

How did you verify your code works?

• Core focused tests: 38/38
• TUI context tests: 4/4
• AI, Core, TUI typechecks: pass
• monorepo typecheck: 33/33 packages
• full Core: 1490 pass, 6 skip; one unrelated WebFetch timeout passed isolated rerun 12/12
• pre-push typecheck: pass

Screenshots / recordings

N/A; no visual layout change.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[github-actions]

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Based on the search results, I found a related PR:

PR #38987: fix(core): honor Codex input limits
#38987

This is the PR that the current PR (39545) explicitly supersedes. According to the description: "Supersedes #38987: same Kit-authored patch, conflict-resolved onto current v2."

PR #39082: fix(opencode): correct Codex OAuth GPT-5.6 context window
#39082

This PR appears related as it also addresses GPT-5.6 context/input window handling in the Codex OAuth flow, which is directly relevant to the current PR's focus on honoring input limits for ChatGPT OAuth.

These are the closest related PRs; PR #38987 is the direct predecessor that was superseded by this current work.

[github-actions]

Thanks for updating your PR! It now meets our contributing guidelines. 👍

[magoz]

Issue link note: the template now contains Closes #38851. GitHub still reports zero closingIssuesReferences because this PR targets v2, not default branch dev (same behavior on #38987), so automation retains needs:issue despite the linked issue.