MMVistaReason: Toward Open-data and post-training recipes for multimodal reasoning