Implementing KG generation pipeline

[thinhngo-x]

The objective of this issue is to implement a complete pipeline for KG generation and to resolve some problems below. After prototyping (PR https://github.com/Viet1004/Code4Earth-2024-Challenge-24/pull/11), I identified some problems:

• Entity resolution: Due to the documents are not input all in once, the generated KGs may have nodes repeated many times. Therefore, Entity resolution is necessary to merge multiple KGs.
• Llama3 has 8K-token context window. It is preferrable to split documents in chunks so that the prompt reach the maximum number of tokens, instead of splitting documents only by pages.

[thinhngo-x]

For counting tokens, keep an eye on this issue on Ollama: https://github.com/ollama/ollama/issues/3582. When this api is added, we can rely on ollama native api to count number of tokens instead of relying on huggingface as in https://github.com/ollama/ollama/issues/1716#issuecomment-2023379127.