Refactor: Migrate legacy AgentExecutor-based LangChain agent to LangGraph

[azamiftikhar1000]

📌 Background / Why

Legacy agents built with AgentExecutor work, but they 

• limit fine‑grained control over decision flows,

• make tool‑calling logic harder to debug, and

• sit on an API path LangChain is actively de‑emphasizing.

LangGraph provides a graph‑structured runtime that delivers:

• Deterministic control‑flow (explicit nodes + edges)

• Easier state inspection & debugging

• Better concurrency & performance hooks

• Alignment with future LangChain releases (long‑term support)

---

🎯 Goals / Acceptance Criteria

• Replace current initialization:

  agent = initialize_agent(
      all_tools,
      llm,
      agent=agent_type,
      verbose=verbose,
      agent_kwargs=default_agent_kwargs,
      **kwargs
  )
  

  …with an equivalent LangGraph graph (e.g., ReAct or custom Supervisor pattern).

• Preserve existing functionality and tool set (all_tools).