feat: ship llm-only turn understanding and memory governance upgrades

Add LLM memory quality eval gate and promote OpenAI-agent readiness updates

Wave 5: add Jarvis persona posture, tone controls, and sim validation suite

Add confidence-aware language policy routing (JX04)

Add context-aware first-response strategy routing (JX03)

Add dynamic brief/normal/deep response mode switching (JX02)

Add style consistency regression suite for common prompts

Implement Jarvis interaction contract prompt enforcement

Handle memory context lookup failures without aborting replies

Clean up test lint debt and dead test variables

Harden style/memory query flow and sanitize non-finite tool inputs

Improve persona style lookup and isolate test memory state

Complete remaining TODOs: attention decay, schema parity, and persona styles

Handle startup failures and persist SDK session ids

Fix sentence boundary streaming and DoA tri-state handling

feat: add tool policy allow deny

feat: add memory sensitivity and task next

feat: add memory store and task plans

feat: add attention mux and gesture macros

feat: initial Jarvis on Reachy Mini