Schrodinger Bridges Beat Diffusion Models on Text-to-Speech Synthesis