feat: add entity profile curation and improve claim review observability

[chenkel-data]

Summary

This PR adds an optional entity profile curation stage after entity resolution and improves how extracted graph claims are tracked, reviewed, and observed.

The new curation workflow stores profile evidence for resolved entities, compares new article evidence against the current profile, and can keep the profile, update it, or flag it for human review. It is policy-aware, so the same article evidence is not reviewed again unless the review or curation prompts change.

The PR also strengthens claim lifecycle handling by tracking active article support, marking claims that lose support as needing review, detecting conflicting directions or competing relationship types, and exposing richer provenance in the frontend.

What Changed

• Added optional AI entity profile curation with review and curation prompts.
• Added profile evidence storage, curation status, history, trace links, and profile embedding refreshes.
• Added MLflow tracing, artifacts, prompt registry links, and workflow-step LLM cost tracking for extraction, gleaning, profile review, and profile curation.
• Refactored the Neo4j graph layer into focused stores for article writes, claims, entity profiles, entity resolution,
  graph reads, insights, and admin actions.
• Improved claim lifecycle tracking with active support counts, unsupported claim review, direction-change handling, and competing-claim detection.
• Updated the frontend details panel with profile curation status, entity extraction traces, cleaner claim review cards, claim filters, and direct MLflow trace links.
• Added graph entity counts to the frontend header.
• Expanded backend and frontend tests for curation, graph stores, ingestion, claim review, and UI behavior.