SCHED-301 Scale metric collection for KSM

[ChessProfessor]

Problem

On large clusters, kube-state-metrics can return a /metrics response larger than the default vmagent scrape size limit, causing scrape failures around ~1.1k nodes.

Solution

Added a configurable max_scrape_size for the kube-state-metrics main http scrape endpoint. The chart default is not raised: by default, kube-state-metrics inherits the existing global vmagent promscrape.maxScrapeSize. Large-cluster deployments can set observability.vmStack.values.kubeStateMetrics.maxScrapeSize explicitly.

The global vmagent limit remains unchanged for other targets.

Testing

Ran:

- helm template test-release helm/soperator-fluxcd
- helm unittest -f 'tests/kube_state_metrics_test.yaml' helm/soperator-fluxcd
- helm unittest helm/soperator-fluxcd
- helm lint helm/soperator-fluxcd
- git diff --check

Release Notes