add clear_gradient、current_step_lr api and parameter_list parameter for optimizer in dygraph (#1694)

* add clear_gradient function and parameter_list parameter for optimizer, test=develop

* fix AdagradOptimizer_cn.rst, test=develop

* add sample code, test=develop

* polish sample code, test=develop

* polish description for clear_gradient, test=develop

* add current_step_lr api for optimizer, test=develop

* polish backward api doc test=develop, test=document_preview

* polish backward api doc test=develop, test=document_preview

* polish doc of optimizer test=develop, test=document_preview

* polish doc of optimizer test=develop, test=document_preview

* polish doc of optimizer test=develop, test=document_preview

* polish code of no_grad_set test=develop

This reverts commit a29dcbec.

* fix the chinese doc for paddle test=develop
paddle.fluid.layers.create_global_var
paddle.fluid.layers.create_parameter
paddle.fluid.optimizer.FtrlOptimizer
paddle.fluid.optimizer.LarsMomentumOptimizer
paddle.fluid.optimizer.MomentumOptimizer
paddle.fluid.optimizer.RMSPropOptimizer
paddle.fluid.optimizer.SGDOptimizer

* polish seven minimize api zh doc

* polish details

* enrich grad_clip param comment