[Features]Improve full-text search for large documents via chunking

[earayu]

The effectiveness of full-text search needs improvement when dealing with very large documents.

Currently, retrieving a massive document (e.g., an entire book) causes two main problems:

1. It can exceed the maximum length accepted by the rerank model.
2. It can overflow the LLM's context window.

To solve this, we should implement document chunking for full-text search. By breaking large documents into smaller, more relevant chunks, we can ensure the retrieved content is more precise and useful for both the reranker and the LLM.