feat: label component pods for status.components

[recharte]

Labels each component's pods so the provider runtime reports them on the Instance: pod counts in status.components (openeverest/openeverest#3335) and the PodsScheduled / PodsReady conditions (openeverest/openeverest#3312).

• RBAC. The runtime needs list and watch on pods to report them. The pods marker was only an example comment, so this adds get, list and watch. make generate syncs the rule to config/rbac and the chart.
• Labels. labelPods puts c.PodLabels(component) on spec.instances[].metadata.labels (engine) and spec.proxy.pgBouncer.metadata.labels (proxy). The operator merges these into the pod templates only. The instance StatefulSet and PgBouncer Deployment selectors keep the operator's own labels, so existing clusters roll once and keep their selectors.

Notes

• go.mod pins core main (da37bb3), which includes feat: report unschedulable and failing pods as Instance conditions openeverest#3312 and the ComponentSpec.name removal from refactor(api): drop ComponentSpec.name openeverest#3337 (no provider used it).
• Existing clusters get one rolling restart when the provider is upgraded, because the pod templates change.

Testing

• make test-unit, make lint (0 issues), make generate.
• Chainsaw core/cluster now asserts the labels on the instance set and PgBouncer specs of the PerconaPGCluster. It also checks that the Instance reports engine and proxy with their selectors and 0/0 pods (the operator is scaled to 0 there). With no pods, the PodsScheduled and PodsReady conditions would be True whatever the runtime did, so they are not asserted; their logic is unit-tested in core.
• Not yet tried on a cluster with the real operator.