patch_all_models() — upgrade MissionController + LearningEngine to Gemma 4

[chatde]

Parent PRD

#1

What to build

Implement patch_all_models() in gemma4_brain.py to transparently monkey-patch mission_controller._ask_vision, mission_controller._ask_text, and learning_engine.VISION_MODEL so all three modules use Gemma 4 without any code changes to those files.

Acceptance criteria

• patch_all_models() replaces mission_controller._ask_vision(pil_image, prompt) → routes to brain.analyze_for_mission()
• patch_all_models() replaces mission_controller._ask_text(prompt) → routes to brain.plan_mission()
• patch_all_models() sets learning_engine.VISION_MODEL = "gemma4:e4b"
• SecurityPatrol mission runs with Gemma 4 (GEMMA4=1 env var) without errors
• ObjectCensus mission produces object list using Gemma 4 vision
• LearningEngine quick scan completes and saves observation to MemoryBank
• patch_all_models() logs confirmation for each patched module via logger

Blocked by

• Blocked by Memory context injection — feed MemoryBank + SpatialMemory into every Gemma 4 call #3 (memory context injection must work first)

User stories addressed

• All Vector missions use Gemma 4 vision — zero changes to existing mission code
• LearningEngine autonomous discovery upgraded to Gemma 4