Summary

• Enable Traefik Prometheus metrics on a private loopback endpoint and have proxy agents forward those samples to VictoriaMetrics with server labels.
• Add agent-side resource metrics for the techulus-cloud-agent process itself: CPU percent, memory percent, and RSS bytes.
• Add service resource metrics for CPU, memory, and network usage, aggregated to stable service_id + server_id series.
• Replace the service overview request-stats log aggregation API/UI with a metrics-only endpoint and chart modes for requests, latency, traffic, and resources.
• Remove the VictoriaLogs request-stat aggregation path and tests.

Verification

• go test ./... in agent
• CI=true pnpm exec tsc --noEmit in web
• CI=true pnpm test -- victoria-metrics-service-metrics.test.ts in web
• CI=true pnpm exec biome check 'app/api/services/[id]/metrics/route.ts' components/service/details/service-details-overview.tsx lib/victoria-logs.ts lib/victoria-metrics.ts tests/victoria-metrics-service-metrics.test.ts in web
• git diff --check -- agent web

Review Notes

• Smart review found two parser issues in agent/internal/container/stats.go around stderr/JSON mixing and name-vs-ID attribution; both were fixed and covered with tests.
• Basecamp review flagged churn-heavy container_id and deployment_id labels on service metrics; those labels were removed and container stats are now aggregated before export.
• Smart review of agent-process metrics flagged CPU percent scale/test coverage; CPU is now normalized to 0-100 like node CPU, and arithmetic guard cases are covered.
• Repo-wide Biome still has unrelated pre-existing diagnostics, so verification was scoped to files changed in this PR.