executor: remove unused /data PVC and dead replicas value from k8s chart

[michaellzc]

Summary

Removes the executor controller's unused /data PersistentVolumeClaim from the Kubernetes-native executor chart (sourcegraph-executor/k8s) — the container volumeMount, the pod-spec volume, the PVC template, and the now-orphaned storageClass / executor.storageSize values / vestigial EXECUTOR_KUBERNETES_PERSISTENCE_VOLUME_NAME env var. Since single-job-pod became the only k8s execution mode (#1163), job pods use their own ephemeral emptyDir at /job and the controller writes nothing to /data. Also removes the non-functional executor.replicas value, since the controller is a singleton that pins Job pods to its own node (replicas > 1 never worked — extra pods contend for the same ReadWriteOnce PVC). RBAC for job PVCs and the separate firecracker/dind chart are intentionally untouched.

Checklist

• Follow the manual testing process
• Update changelog
• Update Kubernetes update doc

Test plan

helm lint passes and all 7 helm unittest cases pass. helm template renders with no mountPath: /data, no PVC volume/resource, and no PERSISTENCE_VOLUME env var.

Validated the upgrade path for an existing installation end-to-end on a local kind cluster:

1. Installed the current main chart (with the /data PVC) → pod 1/1 Running, PVC sg-executor-codeintel Bound (10Gi RWO), Deployment mounting it at /data.
2. Ran helm upgrade --wait to this branch's chart. Result:
   • Helm release deployed (rev 2), rollout succeeded, new pod 1/1 Running with 0 restarts.
   • PVC and its backing PV deleted cleanly (reclaimPolicy Delete GC'd the PV); the old pod terminated so the PVC finalizer cleared within seconds — no stuck Terminating resources.
   • Deployment volumes/volumeMounts empty; ConfigMap env var removed; no Warning events.

Note: on clusters using a Retain PV reclaim policy, the released PV remains and an admin would delete it manually (the /data volume held only lost+found in k8s mode, so no data loss).

Follow-up (not in this PR): the "Job scheduling on Kubernetes" self-hosted docs page still states jobs must co-locate with the executor to share a Persistent Volume — that requirement died with multi-job mode and should be corrected.

[marcleblanc2]

LGTM.

I'm not familiar with the KUBERNETES_JOB_VOLUME_TYPE=pvc mode, but if it's not in use, then we should also remove the PVC permissions from the Kubernetes RBAC role here. DS thread here.