Use Memory Mapping for Model Loading

[AreTaj]

Update forecasting/inference.py to use mmap_mode='r' in joblib.load() calls. This will allow zero-copy loading of models, reducing memory overhead and potentially improving startup time.