v0.2.4 — Clean report output and versioned documentation

This patch removes trailing empty <answer></answer> elements from report.txt and model_response.json, including whitespace-only elements. Reasoning and meaningful answer content are preserved. A response containing only an empty answer element now follows the existing empty-response failure path.

Usage and synthetic-walkthrough links are pinned to this release so they match the installed CLI. The published patch supports chest and abdomen CT; centered cropping and follow-up history remain development-branch features.

Install the release wheel:

uv tool install --python 3.12 https://github.com/MedMLX/MLX-Reason-CT/releases/download/v0.2.4/mlx_reason_ct-0.2.4-py3-none-any.whl

Validation: make env, make qa (119 tests, lint and strict type checks), make build, and a fresh-wheel synthetic FP32 inference run. The model reached EOS without truncation and saved clean, matching text/JSON reports. Synthetic execution does not establish report accuracy or clinical validation.

This is a focused patch to v0.2.3 and retains its public medmlx-core==0.1.2 dependency. The release includes the wheel, source archive and SHA-256 checksums. The pinned weights, model-material licenses and FP32 default are unchanged. Inference requires macOS arm64 on Apple Silicon with Metal.