Added:
- ModalityFusionWithAttention: New module to fuse text, numerical, categorical, temporal, metadata, and graph features using attention.
- EmbeddingPrecomputeWorker: Background worker for periodic precomputation and caching of item embeddings.
Changed:
- Updated ContentEncoder and _compute_embedding in the RecommendationEngine to work with the new fused modality features (input dimension reduced to 256).
- Adjusted model initialization, training routines, and optimizer settings to reflect the new architecture.
- Added minor clarification on Plex URL