3DCF-Labs/doc2dataset

# Technical Summary: 3DCF / doc2dataset Ingests diverse document formats (PDF, Markdown, HTML, CSV, JSON, TeX) into a normalized three-file index (`documents.jsonl`, `pages.jsonl`, `cells.jsonl`) with deterministic macro-cell extraction, then generates task-specific datasets (QA, summary, RAG) with per-cell NumGuard hashes for numeric corruption detection. The Rust core achieves 3–6× token compression versus baseline extractors while maintaining QA accuracy and numeric faithfulness, with exports targeting HuggingFace, LLaMA-Factory, Axolotl, OpenAI finetune, and custom RAG stacks. Provides CLI, HTTP service with UI, and Python/Node bindings; the encoder/serializer is published as `three-dcf-core` on crates.io for library integration