Fixes for concurrent document indexing and querying

[danielaskdd]

Problems Addressed

1. When enable_llm_cache_for_entity_extract is True and enable_llm_cache is False, there is race condition when doing document indexing and user query at the same time. Func extract_entities temporary modify global config enable_llm_cache to True, causing user query also using cache temporary.
2. Performance bottleneck during document indexing due to LLM function instance exhaustion, causing user query latency. The limit_async_func_call decorator implementation using sleep to wait for another thread to release resource is so bad, that user query will not have a chance to run until document index is over.
3. Thread safety issues in NanoVectorDB's disk persistence operations.

Solutions

1. Entity Extraction Improvements:

• Intruduce force_llm_cache parm to handle_cache, forcing the function handle cache no mater what global config settings are.
• Modify extract_entities to use this new parm to control the behavior of handle_cache.

2. Concurrent Operation Optimization:

• Replaced function call limiting mechanism of limit_async_func_call with asyncio.Semaphore, ensuring FIFO order to prevent query starvation during indexing.
• Removed redundant semaphore logic from EmbeddingFunc

3. Introduce asyncio.Lock to protect file save operations in NanoVectorDBStorage

Above changes improve concurrent document indexing and querying reliability while maintaining system performance and improve thread safety.

Bug fixes

1. Fix quantize_embedding for not support embedding vector of list type
2. Fix cache_type missing in llm_response_cache, and add new cache_type extract for entities extraction llm respond cache.
3. Fix prompt respond cache fail when is_embedding_cache_enabled is true.
4. Fix llm_model_func retrieval error in handle_cache, this function is for LLM similarity verification.
5. Update similarity_check prompt to avoid generating two scores sometimes.

[danielaskdd]

@ParisNeo This update involves some core functionalities in LightRAG, and I hope you could help with a preliminary review of the PR.