Fake News in Sheep's Clothing: Robust Fake News Detection Against LLM-Empowered Style Attacks