P2VA: Converting Persona Descriptions into Voice Attributes for Fair and Controllable Text-to-Speech