Move the import of langchain_text_splitters into chunking.default_chunk_fn

[jun76]

Description

• Prevent VoyageAI from pulling in the LangChain stack at import time by moving the langchain_text_splitters import inside the chunking helper.
• Surface a error message if chunking is used without the optional dependency.
• No changes to functional behavior; just defers the heavy dependency until the chunking utilities are actually called.

Background

Thank you for your continued support.

I noticed that simply import voyageai in my project causes python -X importtime to take about 5 seconds.

Upon investigating the cause, I discovered that langchain_text_splitters, loaded in the global scope of voyageai/chunking.py, finds and imports heavy dependencies within my project, such as sentence-transformers, torch, transformers, and scipy.

Therefore, I would greatly appreciate it if langchain_text_splitters could be deferred. I have submitted this PR accordingly.

Thank you.

Pytest

The following has been verified to pass.

pytest tests/test_client.py -k 'with_chunking_fn'