feat(settings): surface quota preflight cutoff in Settings → Routing

[adivekar-utexas]

Summary

The quota-preflight cutoff (resilience.quotaPreflight) already exists in the runtime (it arms credential selection to skip accounts whose remaining quota has hit a cutoff without sending the request upstream), but it had no UI surface — and the /api/resilience PATCH schema silently rejected the section (.strict() dropped it), so it could only be enabled via env var.

This PR surfaces it in Settings → Routing as a "Quota Preflight Cutoff" card:

• Toggle to enable/disable the preflight cutoff globally.
• "Global cutoff (min remaining %)" number field (default 2% remaining = stop at 98% used).
• Per-window cutoffs configured on the Provider Quota page (the Cutoff modal per connection, quotaWindowThresholds) and per-provider window defaults keep precedence; the global default only applies to windows without their own cutoff.

What changed

• src/shared/validation/schemas/settings.ts — added quotaPreflightSettingsSchema and wired it into updateResilienceSchema (previously the section was dropped by .strict()).
• src/app/api/resilience/route.ts — GET/PATCH now round-trip quotaPreflight (PATCH was already computing it via mergeResilienceSettings; it just wasn't accepted on input or returned).
• src/app/(dashboard)/dashboard/settings/components/QuotaPreflightCard.tsx — new card (loads from GET /api/resilience, saves via PATCH /api/resilience with { quotaPreflight }).
• src/app/(dashboard)/dashboard/settings/routing/page.tsx — mounts the card under the existing Routing Strategy card.
• src/i18n/messages/en.json — keys for the card (other locales fall back to English via the existing tx() pattern used elsewhere in the settings UI).
• tests/unit/quota-preflight-routing-ui.test.ts — 3 tests: schema accepts the section, schema rejects malformed values, persisted enabled: true resolves through resolveResilienceSettings (the flag auth.ts reads to arm the preflight latency gate).

Behavior notes

• No default behavior change: the setting remains off by default (existing enabled: false factory default; enabling via env var QUOTA_PREFLIGHT_CUTOFF_ENABLED still works).
• Once enabled, credential selection (getProviderCredentialsWithQuotaPreflight) checks each candidate account's quota windows (5h / weekly / monthly / credits, whatever the provider's quota fetcher exposes — e.g. Command Code's five_hour / weekly / credits) against the cutoff and skips exhausted accounts, falling back to the next account in the strategy order without making an upstream request to the skipped account. This is exactly the "skip out-of-quota accounts in priority order without wasting time" behavior.
• If all accounts are at/below the cutoff, the request returns 429 allRateLimited with the earliest reset time instead of dispatching to an exhausted account.

Test plan

• tests/unit/quota-preflight-routing-ui.test.ts — 3/3 pass.
• tests/unit/sse-auth.test.ts — 66/66 pass (preflight gating + sibling selection).
• tests/unit/combo-priority-quota-exhaustion-cutoff-5923.test.ts, tests/unit/quota-exhaustion-cutoff-opencode.test.ts, tests/unit/issue-6686-quota-preflight-coverage.test.ts — 9/9 pass.
• npx tsc --noEmit -p tsconfig.typecheck-core.json — clean.
• npx eslint on the new/changed non-barrel files — clean (pre-existing barrel-level unused-import errors in schemas/settings.ts are untouched).

[diegosouzapw]

@adivekar-utexas — confirmei que #12090 é esta PR + 3 commits extras (refactor do context_length routing) na mesma branch/autor (git merge-base --is-ancestor confirma que todo commit desta PR está contido em #12090). Fechando esta em favor de #12090, que cobre tudo aqui e mais. Obrigado!