Distributed Saddle-Point Problems Under Similarity