X’s algorithm may turn argumentative replies into a feedback loop that serves holders more content that clashes with their values, with the effect stronger among Democrats, researchers found.