[Safety] 未成年人保护 + 记忆隐私过滤（危机干预已完成） #3

Current limitation:

The project needs a stronger safety layer for high-risk user messages.

Goal:

• Detect self-harm crisis language
• Detect minor-risk contexts
• Avoid romantic / adult interaction with minors
• Add privacy-risk filtering
• Prevent sensitive secrets from entering long-term memory