v0.1.0 — Evidence-grounded MVP

First portfolio-ready release of Review Decision Copilot.

Highlights:

• VADER and optional RoBERTa sentiment analysis across six product aspects
• multilingual embedding + BM25 retrieval with metadata filters
• deterministic or structured LLM decisions with citation validation and abstention
• reproducible retrieval benchmark, Streamlit demo, Dockerfile, and passing CI

The included benchmark uses synthetic public data and is documented as a smoke test, not a production-quality claim.