Redirecting to FlowInOne's Vision-Only Paradigm Could Finally Break Multimodal AI's Alignment Bottleneck...