Learning rule influences recurrent network representations but not attractor structure in decision-making tasks