fix(limit-warmup): repair typing for warmup sender

[Komzpa]

Supersedes #773 because that contributor branch has maintainer edits disabled, while its current head fails ty.

What changed:

• rebases the limit warm-up trigger onto current main
• builds ResponsesRequest via model validation so max_output_tokens is accepted through the schema
• narrows LimitWarmupService constructor dependencies to structural protocols so unit-test fakes type-check
• keeps the request-log protocol aligned with the concrete repository signature
• compares limit_warmup_min_available_percent against available quota, not used quota
• defaults legacy frontend settings payloads for the new limit warm-up fields

Local verification:

• uv run ruff check app/modules/limit_warmup/service.py tests/unit/test_limit_warmup.py
• uv run ty check
• uv run pytest tests/unit/test_limit_warmup.py -q
• bun run typecheck
• bun run test src/features/settings/schemas.test.ts --run

Fixes the CI failure from #773.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3fdef673d5

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".

[Soju06]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ada28b5f80

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".

[Soju06]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 👍

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".

[Soju06]

Hermes review — blockers found

PR #778 is not merge-ready yet.

Blockers

1. Warm-up opt-in reload is still stale within the same SQLAlchemy session.
   app/core/usage/refresh_scheduler.py:82 loads accounts, refresh_accounts(...) runs, then refresh_scheduler.py:94 calls accounts_repo.list_accounts() again before warm-up. But app/modules/accounts/repository.py:44-46 is a plain select(Account) using the same AsyncSession; already-loaded Account identities are returned from the session identity map without populate_existing, expire_all, or a fresh session. So if an operator disables limit_warmup_enabled while the refresh cycle is in flight, LimitWarmupService can still see the old value at app/modules/limit_warmup/service.py:229-233 and send a warm-up request.
   Suggested fix: force-refresh the account rows/flags before warm-up (populate_existing=True, session.expire_all(), or a fresh session/query dedicated to warm-up candidates) and add a regression test for the same-session stale identity-map case.

2. OpenSpec change is structurally invalid.
   uv run openspec validate add-limit-warmup-trigger --strict fails because all three change spec files lack delta sections:

   • openspec/changes/add-limit-warmup-trigger/specs/database-migrations/spec.md
   • openspec/changes/add-limit-warmup-trigger/specs/frontend-architecture/spec.md
   • openspec/changes/add-limit-warmup-trigger/specs/usage-refresh-policy/spec.md

3. Live GitHub mergeability is blocked.
   GitHub currently reports mergeable=false, mergeable_state=dirty, mergeable=CONFLICTING / mergeStateStatus=DIRTY.

Additional non-blocking cleanup worth considering

• frontend/src/features/settings/components/routing-settings.tsx:143-147 adds a Switch with no accessible name. The visible text is not programmatically associated with the switch.
• frontend/src/features/dashboard/components/account-card.tsx:125-139 exposes the per-account warm-up toggle as only “On”/“Off”, ambiguous for screen readers across multiple account cards.
• Frontend warm-up model/prompt validation only checks non-empty (frontend/src/features/settings/schemas.ts:38-39), while backend rejects model >128 and prompt >512 (app/modules/settings/schemas.py:48-49).
• Cooldown parsing uses Number.parseInt (routing-settings.tsx:38-47), so 60.5 silently becomes 60 before save.

Verification run

• git diff --check origin/main...HEAD ✅
• uv run pytest tests/unit/test_limit_warmup.py tests/integration/test_settings_api.py tests/integration/test_accounts_api.py -q ✅ (15 passed)
• uv run openspec validate --specs ✅ (21 passed)
• uv run openspec validate add-limit-warmup-trigger --strict ❌ as above
• Live checks are green, but mergeability is currently dirty/conflicting.

[Soju06]

Addressed the Hermes blockers and related cleanup:

• Merged current main into the PR branch to resolve the dirty/conflicting state.
• Added AccountsRepository.list_accounts(refresh_existing=True) using SQLAlchemy populate_existing and switched the warm-up scheduler to use it immediately before candidate evaluation.
• Added a same-session identity-map regression test so stale limit_warmup_enabled values cannot regress silently.
• Converted add-limit-warmup-trigger OpenSpec files to strict-valid delta specs.
• Tightened frontend warm-up validation to match backend max lengths, rejected decimal cooldown values without truncation, and added descriptive accessible names for warm-up controls.

Local verification:

• uv run ruff check app/core/usage/refresh_scheduler.py app/modules/accounts/repository.py tests/integration/test_accounts_repository.py tests/unit/test_proxy_load_balancer_refresh.py
• uv run ruff format --check app/core/usage/refresh_scheduler.py app/modules/accounts/repository.py tests/integration/test_accounts_repository.py tests/unit/test_proxy_load_balancer_refresh.py
• uv run pytest tests/integration/test_accounts_repository.py tests/unit/test_limit_warmup.py tests/integration/test_settings_api.py tests/integration/test_accounts_api.py -q
• uv run pytest -q (2750 passed, 7 skipped)
• uv run ty check
• uv run openspec validate add-limit-warmup-trigger --strict
• uv run openspec validate --specs
• npx --yes bun@1.3.14 install --frozen-lockfile
• npx --yes bun@1.3.14 run test src/features/settings/schemas.test.ts src/features/settings/components/routing-settings.test.tsx src/features/dashboard/components/account-cards.test.tsx --run
• npx --yes bun@1.3.14 run lint
• npx --yes bun@1.3.14 run build

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Hooray!

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".