* fix docs, test=develop

* fix LoDTensor ref bug, test=develop

* follow lanxianghit's comments, test=develop

* add range of output random int tensor, test=develop