[Bug]: Qwen3-Next NVFP4 quants silently produce garbage when linear_attn weights are missing from quantization_config.ignore