Manipulation Risks in Explainable AI: The Implications of the Disagreement Problem