[codex] Separate RAG flow collection and status handling

[Parshan-MHB]

Summary

This updates APRAG-Lab so the three benchmark flows are separated at both the data-collection layer and final-answer layer, and fixes queue/status handling around uploads and benchmark runs.

Changes

• Blocks benchmark creation while source extraction is still queued or processing.
• Adds real job status to run history responses and frontend display.
• Prevents source-progress polling from restarting on every refreshed source array.
• Separates answer synthesis by flow:
  • Traditional: retrieval-first.
  • Agentic: specialist/critic-first.
  • Hybrid Graph: graph-relationship-first.
• Separates collection strategies by flow:
  • Traditional: fusion hybrid search.
  • Agentic: bounded agent tool collection.
  • Hybrid Graph: graph-first relationship expansion with vector backfill.
• Adds regression coverage for distinct answers, distinct collection strategies, queue blocking, and frontend disabled-run behavior.

Validation

• docker compose run --rm api pytest tests/test_api.py::test_project_upload_and_run_text_vertical_slice tests/test_api.py::test_agentic_pipeline_respects_limits_and_persists_independently tests/test_api.py::test_agentic_trace_contains_orchestrator_packages_critic_and_final_answer tests/test_api.py::test_hybrid_pipeline_answer_citations_and_trace tests/test_api.py::test_epic19_advanced_rag_variants_are_activated_and_capped tests/test_api.py::test_queued_benchmark_is_blocked_until_source_upload_finishes tests/test_api.py::test_epic28_queued_benchmark_is_processed_by_worker tests/test_api.py::test_epic28_source_upload_is_staged_and_processed_by_worker
• docker compose run --rm web npm test -- --run src/test/App.test.jsx

Note: the focused API test run passed; one-off test containers logged expected transient trace-export warnings because Jaeger/Phoenix were not started for the one-off run.