Disentangling AI Alignment: A Structured Taxonomy Beyond Safety and Ethics