Robust inference from multiple test statistics via permutations: a better alternative to the single test statistic approach for randomized trials