Open-Ended Speaking Style Modeling via Fine-Grained and Multi-Granular Contrastive Language-Speech Pre-training