[codex] document G1 architecture analysis and agentic proposal

[raulbastidas1203]

Summary

• document the current real Unitree G1 architecture in DimOS
• propose a modular agentic G1 architecture with deterministic safety boundaries
• compare perception, SLAM, memory, navigation, voice, control, and observability options
• define contracts, diagrams, traceability matrices, an MVP roadmap, and a reproducible evaluation plan

Why

The existing G1 capabilities are distributed across separate blueprints and subsystems. This documentation establishes the verified baseline, identifies integration gaps, and provides a measurable architecture for future implementation without exposing motor control directly to the LLM.

Validation

• verified timestamps in all Markdown files
• checked relative links and balanced fenced code blocks
• checked 67 primary external source links
• confirmed all changes are documentation-only
• did not run robot blueprints or hardware commands