Feature Request: Semantically Chunk Code Files

### Requested feature

For many RAG applications, document processing is the most common use case. In multi-source RAG setups, code repositories often serve as a key knowledge source.

I'm requesting support for processing code files based on Abstract Syntax Trees (ASTs). Instead of breaking chunks at a character threshold, the system would produce semantic chunks, one function or class per chunk (unless it exceeds a configurable length threshold).

A possible implementation could leverage [tree-sitter](https://tree-sitter.github.io/tree-sitter/). Some existing packages that demonstrate this approach include:
- [chonkie](https://github.com/chonkie-inc/chonkie)
- [chunkhound](https://github.com/chunkhound/chunkhound)