* fix one ps gradient clip

* test=develop, add distributed_infer

* add tensor table

* fix ut

Change-Id: I151e152919a1863db07792bffb42d0ca68995756

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

* fix test_dist_fleet_heter_ctr & peformance update

* test=develop, bug fix

* test=develop, add ps auto

* add print test_dist UT elapsed time

* UT elapsed time optimize

* add save/load for parameter server

* support ps-gpu

* move paddle.fleet to paddle.distributed.fleet

* test=develop,test=document_fix, remove the out args

* fleet_util move to paddle.fleet
Co-authored-by: NWuHaobo <wuhaobo1994@gmail.com>
Co-authored-by: Ntangwei12 <tangwei12@baidu.com>

Integrated Trainer of Parameter Server (API add `fluid.contrib.layers.sparse_embedding` only) (#22957)

* Integrated Trainer of Parameter Server

This reverts commit 66fce9e8.

Fleet Parameter Server API Integrated

* add thread barrier for the compiled program

* add half_async in the communicator
* fix DistributedStrategy

* fix grad clip， clip op belongs to Backward op when running in Parameter Server mode.

* add distributed_strategy

add executor.FetchHandler for train/infer from the dataset

* refector geo sgd & communicator

* add print log for unittest of distributed training
test=develop

* fix save/load in Fleet
* add UT framework of Fleet