[fix] Add FoundationDB chart retention guardrails

[Al-assad]

Summary

• keep the FoundationDBCluster resource by default in the doris-foundationdb chart with helm.sh/resource-policy: keep
• document that FoundationDB should be managed in its own Helm or Flux release, separate from Doris config-only rollouts
• document that production metadata volumes should use a StorageClass with reclaimPolicy=Retain

Motivation

FoundationDB stores metadata for storage-compute decoupled Doris deployments. A Doris config-only change, such as updating BE config maps or compute-group settings, should not implicitly tear down the FoundationDB metadata cluster.

This change adds chart-level guardrails to reduce accidental metadata loss when Helm would otherwise delete the FoundationDBCluster resource during uninstall, upgrade, or rollback workflows.

Important behavior

helm.sh/resource-policy: keep prevents Helm from deleting the FoundationDBCluster resource, but the retained resource becomes orphaned from the Helm release. Operators must account for that behavior during uninstall, reinstall, or helm install --replace workflows.

This guardrail does not replace deployment-level lifecycle separation. FoundationDB should still be deployed and upgraded independently from Doris config-only changes.

This guardrail also does not control PV reclaim behavior. Production deployments should explicitly configure a StorageClass with reclaimPolicy=Retain for FoundationDB metadata volumes.