feat: ONNX embedder with model download and SHA256 verification

Add embedder module using ort (ONNX Runtime) for sentence-transformers
all-MiniLM-L6-v2 model inference with HuggingFace tokenizer. Includes
download with progress bar and retry, SHA256 verification, mean pooling
over token embeddings, and L2 normalization.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: devwhodevs