fix(observability): resolve kube-ovn dashboard datasource references

[gurinderu]

Problem

Every panel on the Kube-OVN Grafana dashboards (Controller / CNI / OVN / OVS / Pinger) shows Datasource ... was not found.

Root cause

The upstream kube-ovn dashboards (dist/monitoring/*-grafana.json, release-1.14) reference their datasource two ways, neither of which resolves in this cluster:

1. A DS_PROMETHEUS dashboard input (${DS_PROMETHEUS} references in panels/targets).
2. A datasource template variable hardcoded to the value prometheus.

This cluster's Prometheus-compatible datasource is VictoriaMetrics (uid victoria-metrics, isDefault: true); there is no datasource named/uid prometheus. The dashboards that work (dotdc k8s-views, node-exporter) leave the datasource variable empty, so Grafana falls back to the default — kube-ovn hardcodes prometheus, so it errors.

Fix (two parts)

• dashboard.yml — add spec.datasources mapping DS_PROMETHEUS → VictoriaMetrics to all 5 GrafanaDashboard CRs. Resolves the ${DS_PROMETHEUS} references (the bulk of panels; 9/11 on CNI, 12+ on Pinger).
• datasource-prometheus-alias.yml (new) — a GrafanaDatasource with uid/name prometheus pointing at the same vmsingle-stack backend, isDefault: false, allowCrossNamespaceImport: true. Satisfies the hardcoded $datasource template variable. Non-default, so it does not affect dashboards that resolve the default datasource.

Verification

• kubectl kustomize flux/apps/kube-system/kube-ovn/observability/ builds clean.
• Part 1 verified live on a mainnet cluster: after applying spec.datasources, unresolved ${DS_PROMETHEUS} references dropped to 0 and now point at VictoriaMetrics; all kube-ovn panel queries return data through the VictoriaMetrics proxy.
• Part 2 (alias) was intentionally not applied directly to prod (GitOps); the mechanism mirrors the already-working dashboards — a valid datasource reference resolves.

Deploy note

Reaches the target environment via the per-env tag move (Flux tag > branch).