cluster-manager: ship the PriorityClass the pool chart's prewarm placeholder needs — a pool release cannot create cluster-scoped objects as the tenant account, so prewarm: true fails the pool install #539

Problem

The gpu-node-pool chart's pool.prewarm (0.4.0) renders a cluster-scoped PriorityClass (value -1000) for its placeholder Job. Pool releases are
delivered by the organisation's tenant ServiceAccount, whose rights are namespaced (Flux multi-tenancy): the release fails InstallFailed on
the PriorityClass and the pool never comes up (create_node_pool {prewarm: true} → release failed 6 s later, no NodeClaim). No negative-priority
class exists on an installation for the Job to reuse.

Root cause

A pool release cannot own a cluster-scoped object, and nothing on the installation ships one the placeholder could name.

Solution

Shipped in #540 → 4.32.0: the connectivity chart renders PriorityClass agent-platform-prewarm-placeholder — value: -1000,
preemptionPolicy: Never, globalDefault: false, a description naming its use — from clusterManager.prewarmPriorityClass {enabled: true, name, value},
mirrored in the meta chart. The class renders with the cluster-manager component, which composes the pool releases and is on in exactly one release
per installation (the platform's): on an installation model serving runs in the serving slice cluster-manager creates beside the pool, so a class gated
on the model-serving switch would arrive after the pool release's Job and leave with the slice. The gpu-node-pool chart names the class through
pool.prewarm.priorityClassName (default: this name) and renders none (giantswarm/gpu-node-pool#15).

Acceptance criteria

• The connectivity chart renders the PriorityClass with the cluster-manager component on; enabled: false renders none; the meta chart
  mirrors the values (verify-mirrored-values, verify-cluster-manager — feat(cluster-manager): the connectivity chart ships the PriorityClass the gpu-node-pool chart's prewarm placeholder runs at — agent-platform-prewarm-placeholder (-1000, Never), rendered with the cluster-manager component, mirrored in the meta chart #540).
• README names the class as the one the pool chart's pool.prewarm.priorityClassName defaults to (root README "Cluster manager", connectivity
  README "The prewarm placeholder's PriorityClass" — feat(cluster-manager): the connectivity chart ships the PriorityClass the gpu-node-pool chart's prewarm placeholder runs at — agent-platform-prewarm-placeholder (-1000, Never), rendered with the cluster-manager component, mirrored in the meta chart #540).
• On an installation, the class exists after the release (evidence:
  platform and connectivity HelmReleases on 4.32.0, PriorityClass agent-platform-prewarm-placeholder -1000/Never owned by the connectivity release).
• A pool created with prewarm: true (gpu-node-pool ≥ the release of prewarm: render no PriorityClass — a pool release is namespaced; take the class name from pool.prewarm.priorityClassName (the platform ships it) gpu-node-pool#15) installs and its placeholder pod is admitted
  (evidence in the trial that follows).