Implement caching for query embeddings

[thoj]

Added caching to store query embeddings, preventing redundant API calls for the same query across multiple files in the store.

When using file search with many files rag_api sends a embeddings query for each file in the store. This is completely unnecessary AFAIK. This pull implements a very simple change to speed up file search queries with many files.

On my test agent in LibreChat with over 40 files, Without cache:

rag_api           | 2025-01-24 00:08:07,012 - httpx - INFO - HTTP Request: POST https://api.openai.com/v1/embeddings "HTTP/1.1 200 OK"
rag_api           | 2025-01-24 00:08:07,476 - httpx - INFO - HTTP Request: POST https://api.openai.com/v1/embeddings "HTTP/1.1 200 OK"
.... x40
rag_api           | 2025-01-24 00:08:24,632 - httpx - INFO - HTTP Request: POST https://api.openai.com/v1/embeddings "HTTP/1.1 200 OK"
rag_api           | 2025-01-24 00:08:24,837 - httpx - INFO - HTTP Request: POST https://api.openai.com/v1/embeddings "HTTP/1.1 200 OK"
rag_api           | 2025-01-24 00:08:24,841 - root - INFO - Request POST http://rag_api:8000/query - 200

With Cache:

rag_api           | 2025-01-24 00:16:28,813 - httpx - INFO - HTTP Request: POST https://api.openai.com/v1/embeddings "HTTP/1.1 200 OK"
rag_api           | 2025-01-24 00:16:28,929 - root - INFO - Request POST http://rag_api:8000/query - 200
rag_api           | 2025-01-24 00:16:28,938 - root - INFO - Request POST http://rag_api:8000/query - 200
rag_api           | 2025-01-24 00:16:28,939 - root - INFO - Request POST http://rag_api:8000/query - 200

Notice only one query to OpenAI. This saves almost 20 seconds.