fix(codex): isolate Spark quota and stabilize quota UI

[xz-dev]

Summary

• Scope hydrated Codex quota exhaustion by requested model so an exhausted GPT-5.3-Codex-Spark window no longer filters normal Codex requests such as codex/gpt-5.5.
• Stabilize OpenAI Codex quota row/column order as: Session, Weekly, GPT-5.3-Codex-Spark Session, GPT-5.3-Codex-Spark Weekly, Banked Reset Credits.
• Preserve bankedResetCredits through provider-limits cache serialization and stale refresh fallback so Banked Reset Credits can render without requiring a manual Refresh Now.

Context

Production on v3.8.44 already had per-window quota_snapshots.is_exhausted from #5941, but lazy hydration still promoted any exhausted snapshot row into connection-level exhaustion. When the Spark weekly window was exhausted and normal Codex windows still had quota, the quota-aware selection path could treat the whole Codex connection as exhausted and return 429 for normal Codex traffic. This keeps the #4293/#5941 per-window model boundary during selection/hydration.

Tests

• node --import tsx/esm --test tests/unit/codex-quota-selection-hydration.test.ts tests/unit/sse-auth.test.ts tests/unit/provider-limits-ui.test.ts tests/unit/provider-columns.test.ts tests/unit/db-provider-limits.test.ts tests/unit/provider-limits-recovery.test.ts
• npm run check:file-size
• npm run check:dead-code
• node node_modules/typescript/bin/tsc --pretty false -p tsconfig.typecheck-core.json
• git diff --check

[gemini-code-assist]

Code Review

This pull request introduces support for Codex-specific quotas, including Spark session/weekly limits and banked reset credits. It updates the UI columns, sorting, and caching mechanisms to handle these new fields, and refactors the credential selection logic to use a request-aware quota exhaustion check (isQuotaExhaustedForRequest) instead of a generic connection-wide check. A critical issue was identified in src/domain/quotaCache.ts where isQuotaExhaustedForRequest uses every instead of some to check for exhausted scoped windows, which would incorrectly allow requests to proceed even if one of the required quota windows is fully exhausted.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

Using every here means the connection will only be considered exhausted for the requested model if all of its scoped windows are exhausted. However, if even a single scoped window (such as session or weekly) is exhausted, any request to that model will be blocked and return a 429. To prevent sending requests to connections with partially exhausted quotas, this should use some instead of every.

Suggested change

	return (
	scopedWindowNames.length > 0 &&
	scopedWindowNames.every(
	(windowName) => getQuotaWindowStatus(connectionId, windowName, 100)?.reachedThreshold
	)
	);
	return (
	scopedWindowNames.length > 0 &&
	scopedWindowNames.some(
	(windowName) => getQuotaWindowStatus(connectionId, windowName, 100)?.reachedThreshold
	)
	);

[diegosouzapw]

Merged into release/v3.8.46 — thank you, @xz-dev! Synced onto the release tip, added the CHANGELOG entry, and verified tests green. 🙌