Add growth-log replay and memory feedback

[sparklkt]

What changed

This PR adds growth-log replay and memory feedback to the offline InnerBrain-Factor prototype.

• adds typed GrowthLogReader support for replaying JSONL growth logs
• adds MemorySignal aggregation for recurring risks, successful factor patterns, repeated evidence gaps, repeated human-judgment triggers, unresolved family conflicts, and useful collision lanes
• feeds memory signals back into disturbance scoring, factor generation, and final value review with lightweight rule-based weighting
• extends the pipeline result to expose current memory signals
• adds new CLI commands: innerbrain replay --limit N and innerbrain summarize-memory --limit N
• adds tests for replay, risk-memory reinforcement, offline success reinforcement, and protection against memory weakening high-risk gating

Why it changed

The prototype previously wrote growth logs but did not reuse them. This PR turns prior runs into lightweight memory signals so the system can reinforce repeated safety boundaries and reuse successful offline reasoning patterns without introducing LLMs, network reasoning, or autonomous external behavior.

Validation

• .venv/bin/pytest -> 16 passed in 0.11s

Safety boundary

• remains offline and rule-based
• no LLM integration
• no network participation in reasoning
• no real external actions
• no bypass of human judgment
• memory feedback does not disable human_judgment_required for networking, self-modification, privileged tools, autonomous execution, or real-world impact