Self-Attention as Distributional Projection: A Unified Interpretation of Transformer Architecture