Causal Concept Embedding Models: Beyond Causal Opacity in Deep Learning