Disable OpenObserve infra-metrics + move collector overrides to the wrapper layer

[ashish1099]

Why

Two things, both rooted in one rule: <app>/charts/ is the vendored upstream subchart and must not carry local customization — a re-vendor silently drops it. Customization belongs in the first-layer wrapper (argocd-helm-charts/openobserve/values.yaml).

1. OpenObserve must not ingest infra metrics. Metrics + alerting live only in kube-prometheus.
2. PR Default openobserve-collector exporter to in-cluster router #72's endpoint fix slipped into the upstream subchart — it would have reverted to OpenObserve Cloud (api.openobserve.ai) on the next re-vendor, silently shipping cluster telemetry off-cluster.

Changes

1. Wrapper overrides (first layer) — openobserve/values.yaml

openobserve-collector:
  exporters:                                  # pinned here so it survives re-vendoring
    otlphttp/openobserve:
      endpoint: http://openobserve-router.openobserve.svc.cluster.local:5080/api/default
    otlphttp/openobserve_k8s_events:
      endpoint: http://openobserve-router.openobserve.svc.cluster.local:5080/api/default
  agent:                                      # logs-only
    receivers: {hostmetrics: null, kubeletstats: null}
    service: {pipelines: {metrics: null}}
  gateway:                                    # traces + k8s events + service map
    receivers: {prometheus: null}
    service:
      pipelines:
        metrics:
          receivers: [servicegraph]
  kube-state-metrics:
    enabled: false

• Agent → logs-only. hostmetrics + kubeletstats dropped (duplicated node-exporter / kubelet-cAdvisor).
• Gateway → traces + k8s events, plus the service map. The prometheus scrape and the otlp metrics receiver are dropped (real infra metrics). The servicegraph connector is KEPT — it's derived from traces (service map / RED metrics), exists nowhere in Prometheus, and is part of the tracing experience. The metrics pipeline is narrowed to receivers: [servicegraph]; the traces pipeline is untouched and still feeds the connector.
• Bundled kube-state-metrics disabled — kube-prometheus already ships KSM.

Uses Helm null-coalescing to delete upstream keys — the convention already used here (e.g. cert-manager.io/issuer: null).

2. Subchart restored to pristine upstream (second layer)

Reverts PR #72's edits to charts/openobserve-collector/: exporter endpoints back to https://api.openobserve.ai/api/default/, chart version back to 0.4.4. Net runtime behaviour is unchanged — the collector still exports to the in-cluster router; the setting just lives in the layer that persists.

Metric coverage is unaffected

Signal	Prometheus source
node/host CPU·mem·disk·net	node-exporter
per-pod/container usage	kubelet /metrics/cadvisor
k8s object state	kube-state-metrics

Wrapper chart bumped 0.13.5 → 0.13.6.