Dockerless Serverless GPUs: How I Built a Fast, Cheap AI SaaS with RunPod Flash