feat(platform): verify Spark object-store interoperability

[zhouning]

Summary

• replace the same-node shared Iceberg warehouse PVC with a cross-node MinIO S3-compatible boundary
• verify Spark create/read/write, schema evolution, snapshots and time travel, then match the result through Gravitino and direct object-store metadata inspection
• add fail-closed contract/evidence validation, Kubernetes rehearsal manifests, runtime inventory, CI coverage and ADR-056

Verified

• live Docker Desktop Kubernetes rehearsal: MinIO on desktop-control-plane; PostgreSQL, Gravitino and Spark on desktop-worker; no Spark/Gravitino warehouse PVC
• required platform suite: 709 passed, 9 warnings
• PostgreSQL integration: 5 passed
• PR route-evaluation contract: 41 passed, 2 warnings
• focused M3-10/platform-truth tests: 28 passed, 4 warnings
• runtime dependency constraints: 4 passed
• all 28 CI contract validators, evidence verify/validate, kubectl kustomize, git diff --check and frontend production build passed

Boundaries

This proves local cross-node S3-compatible interoperability inside one Docker Desktop host and cluster. Production object storage, protected workload identity, OIDC/TLS, commit failure injection, cancel/reconcile/lineage, Flink and complete Spark conformance remain unverified and fail closed.

Stacked on #23. This PR must not be merged before its base.