Drop per-segment cards and plain text so only colored word chips wrap in a single container while transcription runs.