HuggingFace Sentence Transformers Embedder

[github-actions]

Description

Add a HuggingFaceEmbedder using the sentence-transformers library for fully local, open-weight embeddings — no API key required.

Motivation

Not every user wants to call an external API. Local embeddings are faster for bulk ingestion, free, and privacy-preserving. all-MiniLM-L6-v2 is a solid default that balances quality and speed.

Acceptance criteria

• Class HuggingFaceEmbedder in ragframework/embeddings/huggingface.py
• Subclasses Embedder from ragframework/base.py
• Configurable model_name (default: "all-MiniLM-L6-v2") and optional device ("cpu" / "cuda")
• Uses sentence-transformers (already in [huggingface] extra in pyproject.toml)
• Import guarded with helpful message; raises EmbedderError on failure
• Unit tests — mock the model or use a tiny test model to avoid large downloads in CI
• Exported from ragframework/embeddings/__init__.py
• CHANGELOG.md updated under [Unreleased]

Files to touch

• ragframework/embeddings/huggingface.py — new file
• ragframework/embeddings/__init__.py — export it
• tests/test_embeddings/test_huggingface.py — new test file

Resources

• Sentence Transformers docs
• Pretrained model list
• RandomEmbedder in ragframework/embeddings/random_embedder.py as reference