Recursive / Semantic Text Chunker

[github-actions]

Description

Add a RecursiveChunker that splits text using a priority list of separators (\n\n, \n, . , ) — falling back to the next separator when a chunk is still too long.

Motivation

The existing FixedSizeChunker splits text at arbitrary character boundaries, which can break mid-sentence or mid-paragraph and degrade retrieval quality. A recursive approach preserves semantic boundaries.

Acceptance criteria

• Class RecursiveChunker in ragframework/document/chunkers.py
• Subclasses TextChunker from ragframework/base.py
• Configurable separators: list[str], chunk_size: int, chunk_overlap: int
• Falls back through separator list until chunks fit within chunk_size
• Unit tests covering: long word (no separator fits), empty doc, single-separator doc, overlap behaviour
• Exported from ragframework/document/__init__.py
• CHANGELOG.md updated under [Unreleased]

Files to touch

• ragframework/document/chunkers.py — add RecursiveChunker
• ragframework/document/__init__.py — export it
• tests/test_document/test_chunkers.py — add tests

Resources

• LangChain's RecursiveCharacterTextSplitter for inspiration
• Existing chunkers in ragframework/document/chunkers.py as reference