feat: Add hybrid quality evaluation system with CLIP and VLM support

- Add FeatureExtractor for CLIP-based image/text feature extraction
- Add ObjectiveMetricsCalculator for technical quality metrics
- Add VLMEvaluator for vision language model evaluation
- Add HybridQualityGate combining objective + VLM evaluation
- Enhance CharacterMemory with visual feature support
- Add quality optional dependency (torch, ftfy, regex)
- Add unit tests for new modules

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

config.example.yaml

@@ -65,3 +65,22 @@ template:
   #   - 1920x1080 (horizontal/landscape): image_film.html, image_full.html, etc.
   # See templates/ directory for all available templates
   default_template: "1080x1920/image_default.html"
+
+# ==================== Quality Control Configuration ====================
+# Configure quality evaluation for generated content
+quality:
+  # Enable quality checking (set to false to skip all quality checks)
+  enable_quality_check: true
+
+  # Hybrid evaluation settings
+  hybrid:
+    enable_clip_score: true           # Use CLIP for image-text matching
+    clip_model: "ViT-B/32"            # CLIP model variant
+    enable_technical_metrics: true    # Use technical quality metrics
+    enable_smart_skip: true           # Skip VLM when objective scores are good
+    smart_skip_threshold: 0.75        # Threshold for smart skip
+
+  # Character consistency settings
+  character:
+    enable_visual_features: true      # Enable CLIP visual features for characters
+    visual_similarity_threshold: 0.75 # Min similarity for character consistency