feat: add monitoring stack (VictoriaMetrics + Grafana)

[disentangle-network]

Summary

Deploy a lightweight monitoring stack on the Always Free OKE cluster using VictoriaMetrics single-node + Grafana.

Research Findings

• Recommended stack: victoria-metrics-k8s-stack Helm chart (VMsingle + vmagent + Grafana + node-exporter + kube-state-metrics)
• Resource footprint: ~550m CPU requests, ~896Mi RAM requests
• Storage: emptyDir with 7-day retention (preserves last PVC slot for future scaling)
• ARM64: All components confirmed arm64-compatible
• Remaining headroom after deploy: ~83% CPU, ~95% RAM

Acceptance Criteria

• Monitoring Helm chart or documented install procedure committed
• Values tuned for Always Free tier (resource limits, emptyDir storage, 7d retention)
• ARM64 images pinned (no :latest tags)
• Dashboard-as-code: cluster overview dashboard provisioned via ConfigMap
• Helm lint and template validation pass in CI
• Documentation: setup and access instructions

Notes

• Protocol-specific metrics (peer count, DAG height, RPC latency) depend on disentangle-node exposing a /metrics endpoint (cross-project dependency with protocol repo)
• If persistent metrics become important later, can upgrade VMsingle to use the last available PVC (50Gi)