feat: ML-based Intent Detection + Affection Point System with Comprehensive Logging

[Afdaan]

🎯 Changes

1. ML-Based Intent Detection (Replace Hardcoded Keywords)

• ✅ Integrated MoritzLaurer/mDeBERTa-v3-base-mnli-xnli untuk zero-shot classification
• ✅ Support bilingual labels (Indonesian + English)
• ✅ Intent caching untuk performance optimization
• ✅ 78% accuracy (11/14 test cases passed)
• ❌ Removed hardcoded regex patterns untuk intent detection

2. Affection Point System

• ✅ Intent → Affection Points mapping (gratitude +5, insult -10, etc.)
• ✅ Emotion → Affection modifier (joy +2, anger -3)
• ✅ Relationship signals (friendliness, conflict) untuk context-aware scoring
• ✅ Auto-calculate relationship level dari affection_points + interaction_count
• ✅ Database persistence via update_affection() method

3. Comprehensive Logging

• ✅ INFO level: Production tracking (affection updates, level changes)
• ✅ DEBUG level: Detailed breakdown (signal calculation, confidence scores)
• ✅ Full traceability dari NLP detection → affection calculation → database save

4. Stats Display Refactor

• ❌ Removed unused positive_interactions / negative_interactions counters
• ✅ Simplified stats output focus pada affection_points dan relationship_level
• ✅ Cleaner, more meaningful user stats display

📊 Files Changed

• config/settings.py - Added zero-shot model config
• core/nlp.py - Implemented ML-based intent detection + logging
• handlers/conversation.py - Enhanced affection calculation + logging
• handlers/response/stats.py - Removed duplicate tracking, simplified output