* add trainer desc config to distributed strategy

* code style modified

* data_feed set lod

* fix bug

* code style

* fix bug