Approximating Multi-objective Knapsack Problems