feat: implement Topic Modeling (LDA/NMF) - Sprint 1 complete

- Created scrapetui/ai/topic_modeling.py with TopicModelingManager class
- Implemented LDA (Latent Dirichlet Allocation) topic modeling
- Implemented NMF (Non-negative Matrix Factorization) topic modeling
- Added category assignment and hierarchical topic structure
- All 16 legacy topic modeling tests passing (100%)

Features:
- Automatic topic extraction with configurable number of topics
- Top words per topic with weights
- Article-to-topic assignments with confidence scores
- Multi-label topic distribution for each article
- Hierarchical topic modeling (parent/child topics)
- Database integration for category assignment
- Comprehensive error handling

Algorithms:
- LDA: Uses CountVectorizer + LatentDirichletAllocation from sklearn
- NMF: Uses TfidfVectorizer + NMF from sklearn
- Both support configurable parameters (iterations, top words, etc.)
- Automatic topic labeling from top 3 words

Test Coverage:
- Empty article lists
- Single article edge case
- Topic assignments and confidence
- Category assignment
- Multi-label classification
- Threshold filtering
- Edge cases (no content, short content, large topic counts)

Export:
- TopicModelingManager exported from scrapetui module
- Backward compatible with existing test suite

Author: doublegate

scrapetui/__init__.py

@@ -43,6 +43,7 @@
 )
 from .database.migrations import run_migrations
 from .config import Config, get_config, reset_config
+from .ai.topic_modeling import TopicModelingManager
 
 
 # Backward-compatible wrapper for migrate_database_to_v2()
@@ -85,7 +86,7 @@ def load_env_file(): pass
 ContentSimilarityManager = None
 KeywordExtractionManager = None
 MultiLevelSummarizationManager = None
-TopicModelingManager = None
+# TopicModelingManager imported from ai.topic_modeling
 EntityRelationshipManager = None
 DuplicateDetectionManager = None
 SummaryQualityManager = None