feat: add iGPT as optional email intelligence retriever

[ShayFeldboy1010]

Summary

• Adds iGPT as an optional alternative to the existing ChromaDB RAG pipeline
• When USE_IGPT=true, product enquiry emails skip construct_rag_queries and call the iGPT Recall API directly — replacing the entire document loading/chunking/embedding/retrieval chain with a single API call
• Fully backward-compatible — default behavior is completely unchanged when USE_IGPT is not set

What changed

File	Change
src/tools/igpt_retriever.py	New module with igpt_retrieve() and igpt_search() functions using the official langchain-igpt SDK
src/nodes.py	Conditional routing + retrieval logic based on USE_IGPT env var
src/graph.py	Added "product related igpt" route to skip query construction
requirements.txt	Added langchain-igpt (optional dep)
README.md	Comparison table, flow diagram, setup instructions, links to docs

Why iGPT

This project uses a full RAG stack to answer product enquiries: load docs → chunk → embed → store in ChromaDB → retrieve → generate. iGPT replaces that entire pipeline with one API call against live Gmail/Outlook data — no local vector store, no embedding model, no create_index.py to maintain.

	ChromaDB RAG (Default)	iGPT
Setup	Load docs, chunk, embed, store, create retriever chain	Set 3 env vars
Data source	Static text files in data/ folder	Live Gmail/Outlook inbox (auto-indexed)
Code	~80 lines across create_index.py + agents.py + nodes.py	~10 lines in igpt_retriever.py

Test plan

• Verify default behavior unchanged: run with USE_IGPT unset, confirm ChromaDB RAG pipeline works as before
• Verify iGPT path: set USE_IGPT=true + IGPT_API_KEY + IGPT_API_USER, confirm product enquiry emails route through iGPT retrieval
• Verify non-product emails (complaints, feedback, unrelated) are unaffected by the flag