fix(usage): account for cached token pricing

[ryanngit]

Summary

• update OpenAI Standard API pricing and GPT fallback ordering for cache-aware estimates
• calculate usage with uncached input, cached input, cache write, output, reasoning, and cache-savings buckets
• preserve cache/reasoning metadata, avoid double-saving streaming usage, and surface breakdowns in usage views
• add focused pricing/cache regression coverage

Verification

• checked current official OpenAI pricing page for Standard API rates
• node --check open-sse/providers/pricing.js
• node --check open-sse/utils/usageTracking.js
• node --check src/lib/db/repos/usageRepo.js
• node --check open-sse/handlers/chatCore/streamingHandler.js
• direct Node pricing/cache assertions passed
• git diff --check
• NEXT_DIST_DIR=.next-upstream-pricing-build NEXT_TRACING_ROOT_MODE=workspace npm run build

Note

• local Vitest config could not run in this checkout because node_modules is symlinked to the live patch repo and does not include vitest; the added test matches the existing Vitest test style and direct Node assertions cover the same pricing cases.