- Restore OpenAI provider support for language models
- Add tenacity retry with exponential backoff for LM and embedding calls
- Per-sample reward computation with parallel asyncio.gather
- New decompose operation (inverse of factorize)
- Regex pattern matching for in_mask/out_mask operations
- Dark/light mode logo for GitHub README
- Upgrade all dependencies (litellm 1.83, openai 2.31, starlette 1.0, etc.)
- Increase test coverage from 73% to 76% (530 tests, up from 434)

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