On the Feasibility of Fully AI-automated Vishing Attacks