Replace langchain text splitters with direct implementations:
- Token chunker: use tiktoken encode/decode directly
- Recursive chunker: implement recursive character splitting
- PDF decoder: use pypdf PdfReader directly instead of langchain PyPDFLoader

Remove langchain, langchain-community, langchain-core, and
langchain-text-splitters from pyproject.toml and Containerfile.flow.