Hub tag base_model:quantized is not a 're-upload' marker

When filtering Hugging Face models to drop quantized re-uploads (GGUF/AWQ copies that duplicate the base card metadata), don't use the auto-generated base_model:quantized:<id> tag. The Hub sets it for any repo that ships quantized weight files, e.g. sentence-transformers/all-MiniLM-L6-v2 gets it because it includes quantized ONNX exports, so the most downloaded original model on the Hub would be dropped. Use format tags (gguf, mlx, awq, gptq, exl2) plus a name-suffix regex like [-_.](gguf|awq|gptq|mlx|exl2|bnb|[2-8]bit|fp8|int[48])\b instead. In the top 20k models by downloads, 9,345 carry the quantized relation, but ~850 of them have no quantized-format tag at all.