[codex] feat: add Codex quota-aware OpenAI scheduling

[chaochaoweb3]

Summary

• Add Codex quota headroom signals to the existing OpenAI advanced scheduler score.
• Use both 5-hour and 7-day Codex usage snapshots so the scheduler can gradually prefer accounts with more remaining short-window and weekly capacity.
• Keep the existing auto-pause, sticky-session, TopK, load, queue, error-rate, and TTFT behavior intact.

Algorithm

For each OpenAI candidate in the load-balance layer, the scheduler now adds two soft score terms:

score += quota_5h * headroom(codex_5h_used_percent)
score += quota_7d * headroom(codex_7d_used_percent)
headroom = 1 - used_percent / 100

• A lower 5h usage percentage receives a higher 5h score, reducing the chance that one account hits the short rolling window too quickly.
• A lower 7d usage percentage receives a higher 7d score, spreading weekly Codex quota consumption across accounts more evenly.
• Missing, reset, or stale quota snapshots are treated as neutral 0.5, reusing the existing reset/staleness guards instead of trusting old usage data.
• Set gateway.openai_ws.scheduler_score_weights.quota_5h or quota_7d to 0 to disable either quota signal.

Why

This is a softer version of quota-aware routing than hard threshold switching: accounts are gradually demoted as their Codex windows fill up, while already-exhausted accounts are still handled by the existing auto-pause logic.

Related: #979, #477, #372

Tests

go test -tags unit ./internal/service -run 'TestOpenAIAccountScheduler_CodexQuotaWeights|TestOpenAIGatewayService_SchedulerWrappersAndDefaults'
go test -tags unit ./internal/config -run 'TestLoadDefaultOpenAIWSConfig|TestValidateConfig_OpenAIWSRules'
go test -tags unit ./internal/service -run 'TestOpenAI.*Scheduler|TestDefaultOpenAIAccountScheduler|TestBuildOpenAIWeightedSelectionOrder|TestOpenAIAccountCandidateHeap|TestClamp01|TestCalcLoadSkewByMoments|TestDeriveOpenAISelectionSeed|TestOpenAISelectionRNG'
go test -tags unit ./internal/config ./internal/service

[github-actions]

All contributors have signed the CLA. ✅
_{Posted by the CLA Assistant Lite bot.}

[chaochaoweb3]

I have read the CLA Document and I hereby sign the CLA