Optimal two-stage log-rank test for randomized phase II clinical trials