fix: prevent storage scheduling priority from preempting database instances #4649

🤖 Generated by the Agentic Engineer

What happened

During a reviewed node rollout, a storage-bound workload preempted database instances. Database failover then waited for their graceful shutdown. The existing storage priority policy is intended to make room by preempting stateless workloads, but ordinary database instances have lower priority too.

Expected outcome

Database instances must not become victims of the storage priority tier. Preserve explicit priority choices, storage-affinity checks, disruption budgets, and graceful failover.

Acceptance criteria

• Add a regression that reproduces the priority mismatch using the shipping admission policy.
• Give only the intended database instances a matching scheduling tier, without elevating them to system-critical priority.
• Prove unrelated pods, explicit priorities, and UPDATE requests are unchanged.
• Pass local tests, exact-head review, and CI; merge through the normal production queue.
• Read back the deployed policy and newly admitted database pods after deployment.

Part of #2043.