fix(kubernetes): co-locate RWO runtime workloads

[MarcStdt]

Problem

One-shot runtime Jobs and the Developer Mode StatefulSet mount the same RWO PVC as the active deployment but had no node constraint. Kubernetes could schedule them on another node, producing a Multi-Attach error and permanently blocking the command queue.

Change

• discover the node of the running pod that mounts the runtime PVC
• pin procedure Jobs, helper Jobs, persistent procedures, and the Developer Mode workload to that node
• leave workloads unconstrained when there is no active PVC consumer

Validation

• go test ./internal/runtime/kubernetes
• go test ./...
• go test -race ./internal/runtime/kubernetes
• production reproduction: Developer Mode init Job was scheduled away from an active RWO PVC and hit FailedAttachVolume

[druid-infra]

Error: This repo is not allowlisted for Atlantis.