feat: align admin monitoring and production usage observability

[fj-sunny]

Summary

• align the admin monitoring dashboard with production Grafana metrics, full-width trend rendering, and real current P95 values for scene generation and evaluation endpoints
• improve official Alibaba Cloud usage ingestion, reconciliation, observability status, billing presentation, and quality governance data
• preserve legacy Web/mobile telemetry compatibility and audio/IELTS behavior while updating production frontend validation

Validation

• frontend/admin: 27 tests passed
• frontend/admin: TypeScript and Vite production build passed
• backend/unispeaking-server: admin test suite passed
• production Docker images for backend and admin built and deployed successfully

[fennoai]

The monitoring, usage reconciliation, telemetry compatibility, and frontend contract changes were reviewed. The admin test slice passes, and the web telemetry checks pass; the local web build could not be run because the workspace has no usable vite executable. One monitoring correctness regression is called out inline.

[fennoai]

[P1] Preserve the Prometheus-backed backend health status

This hard-codes the summary status to UP, so the overview and OverviewPage will continue to show the backend as healthy even when Prometheus reports up == 0 or cannot be reached. The previous implementation derived this field from the up{instance="backend:8080"} metric, and the dashboard still uses this value to choose the normal/error state. Keep the health query (or an equivalent backend health check) here and report DOWN when it fails.

[HansonL622]

LGTM