v0.1.0: Approximate replay with empirical tolerances

End-to-end commit-and-audit protocol for open-weight LLM inference.
Working pipeline for Llama 8B INT8 (strong tier with score witnessing
and anchoring). Qwen 7B INT8 functional but anchor gap unresolved.

This release marks the last milestone before the deterministic GEMM
kernel rewrite, which will replace tolerance-based verification with
exact integer arithmetic.