vector load_documents in batches

[dayesouza]

This pull request refactors the document loading process for several vector store backends to support batch operations, improving efficiency and consistency. The main change replaces the single-document insert method with a batch-oriented load_documents method across all relevant classes, and updates the embedding operation and its tests to use this new batching logic.

Batch document loading refactor:

• Replaced the insert method with a load_documents method in azure_ai_search.py, cosmosdb.py, and lancedb.py, enabling batch uploads for Azure AI Search and LanceDB, and iterative uploads for CosmosDB [1] [2] [3].
• Updated the abstract base class in vector_store.py to define load_documents as the primary method for document insertion, with insert now delegating to it for single documents.

Embedding operation and test updates:

• Changed the batching logic in embed_text.py to flush the buffer when it exceeds batch_size * 4 instead of just batch_size, increasing batch efficiency [1] [2].
• Updated the corresponding test in test_embed_text.py to reflect the new batching behavior, verifying that documents are processed in larger batches and adjusting assertions accordingly [1] [2] [3].