OpenVLThinker: An Early Exploration to Complex Vision-Language Reasoning via Iterative Self-Improvement