Add self-reflection capability to curator agent

[Theadd]

Curator agent should capture insights from its own review process to avoid repetition in future sessions.

Currently: Curator reviews OTHER agents' insights but doesn't capture its own learnings.
Problem: Users repeat similar review decisions across sessions (e.g., 'reject as duplicate').

Solution: Enable curator to record patterns, efficiency improvements, and common issues discovered during reviews. Load these insights next session to inform decisions.

Examples:

• Duplicate detection patterns (what's already in hooks/protocols)
• Review efficiency improvements
• Common insight types and their handling