Migrate to latest LangGraph/LangChain: fix deprecated APIs/imports and unblock dependency upgrades

[ansh-info]

Summary

LangGraph/LangChain have evolved significantly and the current codebase relies on deprecated APIs and imports. The app currently only runs against pinned legacy versions. We need a structured migration to the latest LangGraph/LangChain stack, update imports/usages, and keep behavior stable.

Current Problems (Concrete Symptoms)

• Deprecated agent factory: create_react_agent is deprecated and moved to langchain.agents.create_agent.
• Deprecated tracer import: langchain.callbacks.tracers.LangChainTracer has moved to langchain_core.tracers.langchain.LangChainTracer.
• Removed chains modules: langchain.chains.* no longer exists in newer LangChain; graphrag_reasoning.py imports langchain.chains.combine_documents and langchain.chains.retrieval which now break with ModuleNotFoundError.
• Text splitter moved: langchain.text_splitter.RecursiveCharacterTextSplitter moved to langchain_text_splitters.
• LangGraph API drift: create_react_agent, ToolNode usage, and state handling differ across versions. The code currently assumes the old prebuilt API.

Impact

• Code only runs with pinned legacy versions of LangChain/LangGraph.
• Upgrading to secure versions breaks runtime imports.
• Pip-audit fails unless we keep legacy versions or ignore CVEs.

Goals

• Migrate to latest supported LangGraph + LangChain versions.
• Remove all deprecated imports and APIs.
• Keep business logic intact.
• Ensure tests and CI remain green.
• Allow dependency upgrades without pinning legacy versions.

Migration Scope (Known Hotspots)

• Agents:
  • talk2biomodels/agents/t2b_agent.py
  • talk2knowledgegraphs/agents/t2kg_agent.py
  • talk2scholars/agents/*
  • talk2cells/agents/scp_agent.py
• Tools:
  • talk2knowledgegraphs/tools/graphrag_reasoning.py (chains & splitters)
• Streamlit apps:
  • app/frontend/utils/streamlit_utils.py
  • app/frontend/streamlit_app_talk2cells.py

Suggested Plan

1. Upgrade dependencies (LangChain/LangGraph stack + related) and regenerate lock.
2. Replace deprecated APIs/imports:
   • create_react_agent → langchain.agents.create_agent
   • langchain.callbacks.tracers.LangChainTracer →
     langchain_core.tracers.langchain.LangChainTracer
   • langchain.chains.* → new retrieval/LLM flow
   • langchain.text_splitter → langchain_text_splitters
3. Refactor ToolNode usage where required by the new API.
4. Run full test suite, fix breakages.
5. Update CI pipelines and remove legacy pinning.

Acceptance Criteria

• App runs on latest LangChain/LangGraph without deprecation warnings or import errors.
• Pip-audit passes without ignoring LangChain CVEs.
• Tests and CI pass at 100% coverage where required.

Notes

This migration should be done in stages to avoid breaking production, and each step should be verified against tests and Streamlit runtime.