feat(storage): add multi-realm RGW support and phase-resume fixes

[htohru-ibm]

Summary

Enhancements to scripts/setup-storage.sh to support multi-realm Ceph
clusters and fix several phase-resume failures.

Changes

Multi-realm RGW support (--rgw-realm)

• Add --rgw-realm REALM option for native-mode clusters that have named
  realms (i.e. not the built-in default realm).
• Rewrite detect_rgw_zone() for reliable zone name resolution:
  • When --rgw-realm is provided, use radosgw-admin zone get --rgw-realm=<realm> to obtain the zone name. zone list's
    default_info field contains a UUID, not the zone name, so it cannot
    be used for this purpose.
  • Redirect all debug/warning output to stderr so that the zone name
    captured via $() is not polluted.
  • In the fallback path, extract the zone name from the zones array
    instead of default_info.

Resolve RGW host short name to IP address

• ceph orch ps returns a short hostname that is typically not resolvable
  from an external tenant cluster.
• After obtaining RGW_HOST from the orchestrator, resolve it to the
  node's public IP address using ceph orch host ls -f json + jq.
  If resolution fails, the short hostname is kept as-is.

Auto-detect RGW protocol (http/https)

• Inspect ceph config dump for ssl_port / ssl_certificate in the
  rgw_frontends config and set RGW_PROTOCOL=https automatically when
  SSL is detected.

Script name correction

• Replace all references to ceph-external-cluster-details-exporter.py
  with create-external-cluster-resources.py, which is the current name
  used by Rook/ODF.

Phase-resume credential restoration (phases 6, 7, 8)

• When restarting from a mid-run phase with --phase N, RGW_ACCESS_KEY,
  RGW_SECRET_KEY, and MAIN_RGW_ENDPOINT could be unset because phase 1
  was skipped. These are now restored from the credentials file written by
  phase 5. If MAIN_RGW_ENDPOINT is missing from the file, it is
  re-detected via ceph orch ps --daemon-type rgw.

Other

• Add MAIN_RGW_ENDPOINT_OVERRIDE env var to allow explicit endpoint
  override without modifying the script.
• Add --connect-timeout 10 --max-time 20 to the backing-bucket curl
  call for reliable timeout behaviour on slow or unresponsive endpoints.
• Delete the state file in cleanup_artifacts() so the script can be
  re-run from scratch after --delete.

Testing

Tested against a native Ceph cluster with:

• Single-realm (default) setup — existing behaviour unchanged
• Multi-realm setup (--rgw-realm realm-ocp1) — correct zone resolution
• Phase 6/7/8 resume after a mid-run failure — credentials restored correctly