ViG: Linear-complexity Visual Sequence Learning with Gated Linear Attention