Avatar Video Agent Runtime (proposed) overview

PROPOSED realtime animated-avatar VIDEO runtime, rendered inside the existing kradle headless-Chromium Jitsi sidecar (package-surface:kradle-jitsi-agent-sidecar). TalkingHead.js (RPM GLB + Three.js) renders the character to a canvas that is captureStream()'d and published as the agent's Jitsi VIDEO track; TTS is published as the AUDIO track through a same-page Web Audio graph, with mouth visemes driven off AudioContext.currentTime (lipsync). Cosmetic animation (expression/posture/gaze) is a realtime FAST PATH to the renderer; consequential visual tool calls (canvas content, screen-share/VNC, external video metadata) are GOVERNED by agent-governance-impl:voice-governance@proposed. Extends the voice runtime. See docs/research/realtime-voice-agent-stack.md Part II.