Added document: Accelerate document indexing and retrieval

[writinwaters]

What problem does this PR solve?

Type of change

• Documentation Update

[JinHai-CN]

Two chapter:

1. Accelerate document indexing

• Use GPU for embedding model
• Disable RAPTOR
• GraphRAG will cost lots of time

2. Accelerate question answer, but not retrieval

• Disable multiple-turn dialog
• Disable reranker
• Disable keyword extraction. (will be enable in this version, check with Kevin)
• Check the 'light' after each answer. Give and detailed description on each phase of the answer
  • Total: including chunk retrieval and answer generation.
  • Check LLM: check if LLM is valid
  • Create retriever: create chunk retriever
  • Binding embedding: binding embedding model for answer
  • Bing LLM: binding LLM for tuning question of multiple turn dialog
  • Tune question: time cost of tuning question for multiple turn dialog
  • Bind reranker: binding reranker model for chunking retrieval
  • Generate keyword: keyword extraction from question by LLM.
  • Retrieval: data chunk retrieval cost
  • Generate answer: answer generation by LLM time cost