* Remove the error in the elementwise op, use the backup mode to calculate

* fix bug of cross_entropy_loss,test=develop
* fix log_softmax and some comment,test=develop

test=develop

test=develop

* Add InitializeGPUAndLoadModel to solve random hang when downloading sparse parameters.
* Update SaveBase to solve test problem.

* Merge hapi into Paddle

Hapi is a high level API for training and inference.
The main modules include Model, Loss, Metrics, Dataset.
Also includes common modules and models in NLP and computer vision, such as BERT, ResNet.

These modules are developed by:
0YuanZhang0, guoshengCS heavengate, LielinJiang, qingqing01, xyzhou-puck huangjun12, wangxiao1021, zhangyang.

* fix rnn check_type list error

* tigger ci, test=develop

* update modify, test=develop

* add new macro BOOST_GET_SAFELY & unittests, test=develop

* add different macro type, test=develop

* fix get macro type in executor, test=develop

* four macro part change backup

* using one macro for all case, test=develop

* revert attribute change, test=develop

* change to three func to solve gcc4.8 bug, test=develop

* polish some details, test=develop

* support save_inference_model in program_translator test=develop

* fix compatibility with OrderedDict.values() in python3 test=develop

* synchronized random_seed test=develop

* Polish Error Message test=develop

* copy dygraph api to paddle.imperative, test=develop

* polish the code, test=develop

* polish code, test=develop

* polish code, test=develop

* move paddle.imperative.Layer to paddle.nn.Layer, test=develop

* [Dy2Stat] Add test for bert pretraining. 

* Construct fake data. 

* Synchronous random seed of program. 

CreateStepScopes in recurrent op also clears scopes, which can cause segmentation fault un multi-threading. We add a lock in this PR but it may slow the computation process. We will fix in another way in next PR.

* fix numpy ndarray mul var base error; test=develop

* add comment for __array_ufunc__ ; test=develop

* move unitest from imperative math op path to test_math_op_patch_var_base;
test=develop

* raise index error when slice out of range; test=develop

* add uni test; test=develop

* fix format error; test=develop

* add comment for py::index_error; test=develop

* polish error message; test=develop

* polish error message; test=develop

* Enabled quantize all and skip missing in QAT

* add clean up; test=develop

* add import; test=develop

* support to train in static

* support to independent decorator

* remove in_dygraph_mode condition in ProgramTranslator

* fix import param_guard and add train/eval test=develop

* Modify into ShareVarsFromScope and rm __all__ in partial_program test=develop

1. To make ProgramTranslator to support `assert` grammar, this PR adds `assert` python API and C++ code. 

2. Fix a bug: graph_pattern_detector.h #include <gtest/gtest_prod.h> but didn't declared dependency at CMakeLists, which can cause single build failure.

3. Refactoring `Formatter` in print_op to make it reusable and reuse the formatter to print in assert op.

test=develop
test=win_gpu

目前在while_loop的执行过程中，loop_vars中的变量在每次的循环中都会进行拷贝，但是LoDTensorArray类型的变量在while循环体中已经完成了读/写的操作，即完成了更新，此时在进行拷贝属于冗余的操作，故该PR跳过每次循环中loop_vars中LoDTensorArray类型的变量的复制过程。

在PaddleCV/ocr_recognition/atention模型的预测过程中进行性能测试：
|性能|with this PR|without this PR|提升|
|---|---|---|---|
|速度|4957.4ms|4978.47ms|0.4%|

* fix error message, test=develop

* add unittest, test=develop

* add UT for mkldnn_cache_capacity

test=develop

* fix comparison of integer expressions of different signedness

test=develop

* API/OP (margin_rank_loss, nce, row_conv, glu, positive_negative_pair) error message enhancement, test=develop

* add unittest for glu, test=develop

* Replace dygraph_to_static_func with @declarative or program_translator.get_func in test_list.py

* Add comments in ConditionalBlock.

* Support list pop last item. 

* Support pop the i-th item. 

* Support an empty tensor array as Input in assign op and set the kernel type is float.