Speech style on E5 — an optional voice_style {stability, style, speed} and instructions checked before any call, each candidate sent the parts its model takes (ElevenLabs voice_settings; OpenAI instructions except tts-1, and a speed other than 1), the top-level speed still refused, and X-Kamo-Speech-Style naming what the answering model applied (KamoCollab CP06 Task 12)