Pavelevich/llm-checker v3.2.8

on GitHub

v3.2.8\n\nHotfix release for multimodal recommendation quality:\n\n- Fixed false multimodal matches caused by noisy input metadata.\n- Hardened modality inference so image flags alone no longer promote coding-only models into multimodal picks.\n- Added deterministic regression coverage for this scenario.\n\nIncludes prior hotfix commit and release prep metadata updates.