v0.2.0 - end-to-end inference system

The first end-to-end release.

Highlights:

• OpenAI-compatible and vLLM backend
• structured Python verification loop
• parallel reasoning families and entropy-aware consensus
• true total deadlines and cooperative cancellation
• JSONL evaluation and privacy-conscious telemetry
• installable CLI, wheel, documentation, security policy, and contributor workflow

This release contains no model weights, competition data, private prompts, or claimed benchmark result.