Concept Embedding Models: Beyond the Accuracy-Explainability Trade-Off