cherry-pick of PR #37566:

Based on #37411, this PR:

    Continue to fix the bugs when bias add is none in static graph for fused_attention op.
    Polish and improve the unittests in test_fused_attention_op_api.py.

Add support for bias is none for fused_attention op.

* Add paddle.incubate.graph_send_recv API

* fix bug in CudaAtomicMin and CudaAtomicMax

* add empty line

修复了fused_transformer_encoder_layer fine-tune过程发现的一些问题：

    fused_attention_op添加attn_mask=None的支持：PR
    pre_layer_norm处理问题：PR
    参数处理，计算错误的问题：PR
    add_bias计算错误问题：PR
    添加pure fp16的支持：PR

* add mlperf optimization PRs

* update

* Fix fused_attention english doc test=document_fix

本PR是fused_transformer的layer层代码，包含FusedFeedForward的layer层代码和FusedTransformerEncoderLayer的代码。

功能：本PR的目标是提高attention模块的计算性能。
为了减少框架层对op的调度开销，本PR通过在C++层手动实现attention模块，对外提供attention 大op；
为了减少防存开销，本PR采取了两种优化方法：
（1）在q,k,v计算时通过共享输入X，将该处的gemm，transpose和bias add从三次调用减少为一次；
（2）使用kernel融合优化技术，在不同cuda kernel之间通过寄存器传输数据；

* add op: fused_feedforward(backward) (#35611)

这个PR是fused_feedforward反向的代码

相关kernel实现：fused_dropout_act_bias, fused_residual_dropout_bias, fused_layernorm_residual_dropout_bias

fused_feedforward是一个融合算子，该算子对transformer模型的feed forward层的算子进行融合和封装，使得前端只呈现一个接口，通过融合减少部分访存和kernel launch的时间，以此提升性能。

* Move fused_attention and fused_feedforward functional api path to incubate (#36704)

将 #35905 和 #35843 PR中新增的的python api接口移到incubate目录下。

* cache core.ops, test=develop

* refine, test=develop

* softmax mask fuse upper triangle

* cover not implemented cpu code

* update 2.0 public api in all left files

* reverse device.py all list;
fix some flake8 errors

* remove fluid for auto_checkpoint.

* fix bug.

* Remove old custom OP to reduce whl package volume

* [Custom OP]Remove old custom OP to reduce whl package volume

* test=develop, fix test_lookahead

* test=develop, add model_average and lookahead

* add + - * / @ [] operator to ComplexVariable, also add unittest

* fix circular reference bug

* fit for py2.7

* remove reverse oprators which not supported now

* refine parameters order && doc

* update rmsprop doc

* refine adam/transpose/unstack/stride_slice

* fix bug && doc

* fix doc

* bug fix

* bug fix

* fix doc

* fix doc

* fix doc

* fix doc

* depercate old strided_slice

* update doc

* set default value for name

* update doc

* Move hapi form paddle/incubate to paddle

* Remove vision/datasets/utils.py and clean code

* Add sample code for conll05

* Print pull path when saving model

* Fix sample code after paramter_list of SGD is changed to parameters

* Fix bug in wmt16 datase

* fix bug

* fix sample code for hapi.model.save, test=document_fix

* test=document_fix

* update usage of 2.0 API, test=document_fix

* fix bugs, return dygraph back to users while using model.save in dygraph

* fix code style

* refine manual seed

* fix ci problem

* fix unittests

* fix unittest

* set is_init_py=false in manual_seed

* fix unittest

* fix bernoulli_op

* fix(unittest): change random_seed to manual_seed

* 🐞fix(unittest): fix manual_seed

* trigger ci

* fix test_sentiment

* fix test_imperative_save_load

* fix test_uniform_random_op

* fix test_uniform_random_op

* fix test_jit_save_load

* merge develop

* fix manual_seed

* fix manual_seed

* use global engine

* use shared_ptr

* fix double free

* fix bug

* fix bug

* fix bug

* fix test bug

* fix test bug

* fix test bug

* fix ci

* add InputSpec

* add unittest for tensorSpec and SimpleNet

* update save_inference_model for hapi

* update save_inference_model to support dygraph

* fix comments

* fix comments

* test=develop

* test, test=develop

* fix dim test, test=develop

* test, test=develop

* add test_export_deploy_model_dynamic

* fix unittest for hapi: save_inference_model

* fix code style

* accept review by guoshengCS

* fix coverage rate

* update doc for save_inference_model and copyright

* change test model back to LeNet() in test_export_deploy_model

* copy jit.save, use LeNet() to test export deploy model

* add return value for dygraph, and fix doc error

* corrected the doc writing

* Delete redundant import and correct import order in sample code.

* remove 'fluid' and add prepare() and fit() in sample code

* correct usage of API 2.0 in sample code

* fix sample code bugs

* fix code style bugs

* fix test_model.py bugs

* set for_inference=True

* correct usage for static.InputSpec

* update doc for model.save

* correct usage of API 2.0

* rename param name for model.save

* correct for_inference as training

* Move paddle/incubate/hapi/metrics to paddle/metric
* Add Precision, Recall and Auc metric

* fix resource warning. test=develop

* update Conv2d Conv3d conv2d conv3d api

* add map_style dataset. test=develop

* Rename `Input` into `InputSpec`

* fix argument place of Input api

* 1.Fix api sum:(1) input->sum; (2)dim->axis; (3)keep_dim->keepdim.

* 2. fix bug when len(axis) == len(x.shape).