Speculative Model Risk in Healthcare AI: Using Storytelling to Surface Unintended Harms