Add provider analytics and refresh controls

[Mister-JP]

Summary

• refresh landing-page starter recommendations at most once every 24 hours, with explicit refresh forcing a cache update
• record provider usage, token consumption, costs, latency, outcomes, and failure causes for every AI request path
• raise model output budgets while preserving reasoning effort and improve incomplete-response handling
• add database migrations, regression tests, documentation, and journey-map/design artifacts

Validation

• npm test (production build and 24 tests)
• npm run typecheck
• npm run architecture:check
• ESLint on changed source files
• clean migration replay and local D1 schema verification