Application of machine learning to quantify forest cover loss in the Congo Basin and its implications for large mammal habitat suitability