* Dygraph Recompute support AMP

* dygraph recompute: update unitest

* fix precision of mp

* fix bug of seed

* fix dp

* print group

avoid polluting logging's root logger

* fix model parallel

* rm parallel_help.py

* add embedding

* Dygraph reocmpute

* unitest for Dygraph reocmpute

* dy recompute remove unitest for win and mac

* add pipeline layer

* add clip/check

* add amp & clip grad in dygraph

* add logging

* support dp & mp

* add PullSparseValue for pull sparse

* fix bug for PullSparseValue

* add test mode in lookuptable

* revert API change

* add comment for is_training

* topo and memory performance for heterps; test=develop
* add trainwithprofiler in heter trainier; test=develop

* test=develop, add distributed_infer

* oneps (3/4)
Co-authored-by: NMrChengmo <cmchengmo@163.com>
Co-authored-by: Nmalin10 <malin10@baidu.com>
Co-authored-by: Nchengmo <chengmo@baidu.com>

* update, test=develop

* test=develop, rm pathlib

* update, test=develop

* add gloo initializer, test=develop

* set default value to strategy in distributed_optimizer test=develop

* add gloo initializer, test=develop

* test=develop, bug fix

* test=develop, util documents

* fix api document

* move paddle.fleet to paddle.distributed.fleet