feat: add entity deduplication after graph building

[Stayfoool]

Hi @666ghj

I noticed that during graph building, Zep sometimes creates duplicate
entity nodes for the same real-world entity (e.g. "特朗普" and "美国总统特朗普"
appear as separate nodes). This affects the accuracy of the knowledge graph.

This PR adds an automatic entity deduplication step after graph building,
using name similarity pre-filtering + type compatibility check + LLM
confirmation to identify and merge duplicates.

Would appreciate it if you could take a look when you have time.
Happy to make any changes based on your feedback. Thanks!

Summary

• Add entity deduplication service that identifies and merges duplicate
  nodes in the knowledge graph after building (e.g. "特朗普" vs "美国总统特朗普")
• When merging duplicate nodes, migrates all edges from removed nodes
  to the primary node before deletion, preserving graph connectivity
• Three-layer filtering: name similarity pre-filter → type compatibility
  check → LLM confirmation
• Integrate into graph build pipeline automatically (80%-90% progress stage)
• Add standalone POST /api/graph/deduplicate endpoint for manual dedup
• Display dedup report in frontend showing which entities were merged

Changes

• New: backend/app/services/entity_deduplicator.py
• Modified: backend/app/api/graph.py
• Modified: backend/app/services/__init__.py
• Modified: frontend/src/views/Process.vue
• Modified: frontend/src/views/MainView.vue
• Modified: frontend/src/components/Step1GraphBuild.vue

Closes #145