Publish the embedding service on the host ports that the harness already polls and uses for retrieval sanity checks. This keeps the follow-on validation branch stacked on top of NVIDIA#1497 without changing harness behavior.

Summary

• expose the embedding service on the host ports that the harness already expects in attach mode
• keep the follow-on validation for PR Sohail/bump nim versions NVIDIA/NeMo-Retriever#1497 stacked and minimal by fixing the compose side instead of changing harness behavior
• confirm uv run nv-ingest-harness-run --case=e2e --dataset=/raid/jioffe/bo20 passes on top of PR Sohail/bump nim versions NVIDIA/NeMo-Retriever#1497 with this change

Test plan

• UV_LINK_MODE=copy uv run nv-ingest-harness-run --case=e2e --dataset=/raid/jioffe/bo20

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.
• If adjusting docker-compose.yaml environment variables have you ensured those are mimicked in the Helm values.yaml file.