[miki3421]

Summary

• access the private SeaweedFS filer through the existing service_endpoint() helper after the Service and NetworkPolicy are applied
• keep in-cluster operator traffic on the Kubernetes Service while using an RBAC-authorized, loopback-only, ephemeral port-forward for out-of-cluster integration runs
• pass the scoped filer endpoint explicitly to the SeaweedFS client and close the tunnel after readiness and bootstrap operations
• remove the obsolete Kind 9090 -> 32091 mapping and fixed localhost:9090 assumptions from the affected integration scripts
• make wait_for_http() raise TimeoutError when its deadline expires, as documented

The SeaweedFS Service remains ClusterIP; this change does not restore any filer or master NodePort and does not relax the NetworkPolicy.

Validation

• python -m unittest discover -s tests -p 'test_*.py' -q — 129 tests passed
• focused private-service and SeaweedFS tests — 29 tests passed
• Python syntax validation for the five affected .ipy integration scripts
• YAML parsing for .github/kind.yaml
• git diff --check

The destructive integration scripts were not run against a shared cluster. The PR CI uses an isolated Kind cluster and will execute the full and slim SeaweedFS/Milvus scenarios.

Fixes apache/openserverless#281