Co-authored-by: Nbaiyangfan <baiyangfan@baidu.com>

Add 3d Parallelism
Co-authored-by: NWangXi <wangxi16@baidu.com>
Co-authored-by: NJZ-LIANG <jianzhongliang10@gmail.com>
Co-authored-by: Nroot <root@yq01-sys-hic-k8s-v100-box-a225-0562.yq01.baidu.com>

* update, test=develop

* fix test_pipeline, test=develop

* add device_worker for pipeline, test=develop

* Refactor code for dump_field & dump_param: abstracting the common function in base class.
* Support dump randomly & random with lineid
* Support specify the random interval, which avoids printing too much logs.

* Add two types of Metric Calculator: MultiTaskCalculator & CmatchRankCalculator.
* Add a config for DynamicAdjustChannelNum function to denote whether we will discard the remaining instances when they are not be distributed evenly.
* Remove CPU code in Pull/PushSparse and we will add it back when testing it fully.
* Fix some known issues: such as copying persistable vars after one epoch running.

cmake选项中添加了WITH_NCCL，显示指定是否编译NCCL的部分代码，WITH_NCCL默认打开，但如果WITH_GPU为OFF，则关闭WITH_NCCL

添加了PADDLE_WITH_NCCL定义

单机单卡能够关闭NCCL编译，多卡的话需要默认打开NCCL，如果关闭NCCL，则只能使用单卡
Co-authored-by: N石晓伟 <39303645+Shixiaowei02@users.noreply.github.com>

* Add a single_process_multi_thread transpiler.
* Add some UTs.
* Fix some API description.

add executor.FetchHandler for train/infer from the dataset

* add a place field in DataFeed to denote which place it will feed data to.
* abstract the copy process in CopyToFeedTensor function
* add UT for float32 type and for CUDAPlace

(1) use channel instead of vector/BlockingQueue in Dataset，to keep same with existing implementation, and make code more readable and flexible (dataset single output channel or multi output channel). one previous memory out of limit problem is cause by not release memory after training.
(2) add Record because MultiSlotType costs too much memory (80B)，fix memory out of limit problem.
(3) add Channel, Archive in paddle/fluid/framework
(4) change dataset from shared_ptr to unique_ptr in pybind
(5) move create/destroy readers from trainer to dataset
(6) move shuffle from datafeed to dataset. dataset holds memory, datafeed is only for load data and feed data to network.
(7) fix thread num bug of Dataset when filelist size < thread num
(8) support set_queue_num in InMemoryDataset

Add Pipeline Concurrency Train Mode:
- Cpp: pipeline_trainer & section_worker
- Python: PipelineOptimizer
- Add a new data_feed type: PrivateInstantDataFeed
- Add a test demo of pipeline trainer and the test model is gnn
- Do not support win32 now