A.I. chatbots posing as therapists that are programmed to reinforce, rather than to challenge, a user’s thinking, could drive vulnerable people to harm themselves or others, the nation’s largest psychological organization warned federal regulators.