fix(providers): admit larger Nous catalogs within native limits

[lidge-jun]

Summary

Nous model discovery can reject a valid paid/free catalog above its old 256 KiB response cap. Raise only the Nous cap to 1 MiB while retaining its 512-row limit and the shared 4 MiB ceiling.

Carry #3939 from e31f5be74af6eb31ad871b22ab09cbace345fdd1, preserving Vocllum. Add isolated synthetic regressions through the real catalog gather and management connection probe: 390 rows above the former cap, expected paid/free IDs, model count, cache publication and reuse. Existing shared overflow/cancellation and raw-row limit tests remain authoritative. No new provider or credential destination is introduced.

Verification

• Local product tests, typecheck, builds and dependency installs: NOT RUN, explicitly prohibited by the owner.
• Main inspected the three-file diff, source registry equivalence, fixture home/credential isolation and injected outbound transport; Git diff checks passed.
• Independent source/security audit PASS, followed by interdiff re-review PASS on 711a3f65c5b5c8398b4309955cc20d8002c4c31c.
• Current-head Cross-platform CI completed successfully. Linux and macOS logs explicitly show the new 390-row gather/cache and probe regressions passing. The privacy gate passed. Windows runtime shards and macOS unsharded control were skipped; they are not counted as passes.
• Initial head d5778ac8d7 failed because the synthetic bearer literals did not match the repository fixture allowance. Four fixture strings were corrected without changing scanner policy; the superseded failing run was cancelled and is not green evidence.
• CI checked synthetic merge caeaf45 of this head into 514350e6f. Later dev change 9c54000c9 removes historical account captures/notes only; runtime, tests, manifests, lockfiles and build inputs are identical. Whole final-tree identity will be checked against the merge calculation; that is distinct from claiming the final docs combination executed in CI.
• Owner-authorized maintainer integration into dev, subject to current-head review/CI and no outstanding maintainer objections. Original author credit must remain in the squash commit.
• The commit retains only this single bug. No release or deployment is included.

Checklist

• Scope stays focused and avoids unrelated cleanup.
• Docs or release notes were updated when needed; this restores existing discovery behavior without a new user setting.
• Security-sensitive changes were reviewed for secrets, auth, and unsafe defaults; tests use isolated synthetic credentials and intercepted transport.

Co-authored-by: Vocllum 149675937+Vocllum@users.noreply.github.com

Pre-merge refresh: current dev 01c23aedcdfcb913151a2ac8f7acebda58d91eee includes another C documentation/capture cleanup only. git diff --quiet 514350e6f origin/dev -- src tests gui scripts package.json bun.lock tsconfig.json .github exits 0. No runtime/test/build-input drift.

Summary by CodeRabbit

• Bug Fixes

  • Improved Nous model discovery for larger catalogs, supporting responses up to 1 MiB.
  • Ensured mixed paid and free model catalogs exceeding 256 KiB are discovered successfully.
  • Improved connectivity checks so larger Nous catalogs can be validated and counted correctly.

• Performance

  • Confirmed that successfully discovered Nous model catalogs are reused from cache, avoiding unnecessary repeat requests.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Team

Run ID: 98b48aff-6641-41ac-a395-ef9f7a1d9f16

📥 Commits

Reviewing files that changed from the base of the PR and between 514350e and 711a3f6.

📒 Files selected for processing (3)

• src/providers/registry.ts
• tests/providers/provider-connection-test.test.ts
• tests/providers/provider-model-discovery-contract.test.ts

---

📝 Walkthrough

Walkthrough

The Nous provider now permits model-discovery responses up to 1 MiB. Tests cover authenticated probing, 390-model discovery, discovery status, live model counts, caching, and repeated gathers.

Changes

Nous catalog capacity

Layer / File(s)	Summary
Response limit and connectivity probe src/providers/registry.ts, tests/providers/provider-connection-test.test.ts	The Nous discovery limit increases from 256 KiB to 1 MiB. The probe test sends a large authenticated catalog request and verifies 390 returned models.
Discovery and cache contract tests/providers/provider-model-discovery-contract.test.ts	The contract test uses temporary credentials and a payload above 256 KiB. It verifies discovery status, the live model count, cached results, and no second fetch during repeated gathering.

Estimated code review effort: 2 (Simple) | ~10 minutes

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch codex/248-a-nous-8c62

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

✅ Deterministic PR hygiene checks passed.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-07T22:52:03.246680Z	711a3f6	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.