Evaluating biases in simulated snow albedo feedback in two generations of climate models