Personalized Lightweight Text-to-Speech: Voice Cloning with Adaptive Structured Pruning