AlphaAvatar v0.6.4 introduces shared perception and online ENV memory

AlphaAvatar v0.6.4 updates the open-source multimodal assistant runtime with a new architecture centered on six main areas: Core, RTC, Agent, Plugins, Tools, and Channels.

• The update implements a shared perception path where an RTC adapter publishes media once for multiple consumers like Persona, Vision, and Memory to process independently.
• It introduces online ENV Memory, which extracts structured environment memory from continuous multimodal context rather than just conversation text.
• The Agent component is separated into parallel runtimes (Session, Context, Perception, IO, Agent) with AvatarEngine as the orchestrator to clarify plugin dependencies.
• LiveKit remains the default RTC integration but is treated strictly as an adapter boundary to allow future native WebRTC support.

This shift positions AlphaAvatar less as a frontend and more as a reusable runtime layer for persistent realtime multimodal agents.