feat: initialize rag based on HugeGraph

[Ling-Yuchen]

close #19
Feature: utilize relevant data in the graph database as context for llm to response to the natural langauge query (RAG).

Steps involved:

1. Extract keywords: extract and expand the keyword set from the query sentence by llm
2. Query graph: get center nodes by matching keywords, extract a sub-graph with a certain deep as "knowledge"
3. Synthesize answer: get answer from llm through the original query and knowledge from graph

All the data required and generated in the whole process is transfered through a dict "context", explained in examples/graph_rag_test.py.

[javeme]

LGTM