TCL-5951: Add DeleteNode to SlurmControlInterface

[jhu-svg]

Summary

• Adds DeleteNode(ctx, nodeset, nodeName) to SlurmControlInterface, enabling the operator to programmatically remove Slurm node registrations
• Previously, scontrol delete was only callable from inside a dying pod's PreStop hook — if that hook doesn't run (force-delete, OOM, node crash), the Slurm entry persists as a ghost forever
• Takes a node name string (not a pod) so callers can delete orphaned entries that have no corresponding pod — needed for the follow-up orphan reconciliation PR
• Follows the established MakeNodeDrain pattern: lookupClient → Get → Delete, with tolerateError on 404/204 for idempotency

Context

Part of TCL-5951 (Slinky operator lifecycle gaps). This is PR 1 of 3:

1. This PR — adds the capability (no behavior change)
2. PR TCL-5951: Reconcile orphaned Slurm node registrations #26 — orphaned node reconciliation (uses DeleteNode to clean up ghosts in the sync loop)
3. TCCO PR #500 — detects PVCs bound to occupied (not just missing) nodes

Test plan

• Unit tests: successful delete (verifies node removed from fake cache) and idempotent delete (non-existent node returns nil)
• go build ./... passes
• Staging E2E verified via PR TCL-5951: Reconcile orphaned Slurm node registrations #26 — DeleteNode was exercised on staging cluster jhu-test-slurm-slinky-gap-orphaned-node (s2-us-central-8a). Operator logs confirmed "deleting slurm node" for injected orphan nodes (slinky-99, slinky-5, slinky-10, slinky-42). Idempotency verified (deleting already-gone nodes returns nil).