Generalizing the Linear Step-up Procedure for False Discovery Rate Control with Applications to Setwise and High Dimensional Variable Selection