cherry-pick #23645

[Cherry-pick 2.0] Error message enhancement of conv2d_fusion and bugfix related to control nodes detecttion in fusion_group (#23985)

* Op(conv2d_fusion) error message enhancement. (#23596)
* Fix the error misjudgment when there are control nodes in graph. (#23943)

* OP error message enhancement of l2_normalize, matmul, mean, etc 

* [Cherry-pick][dy2static]Support recursive call (#23900)
  * Support recursive call.

  * Remove Redundant decorator to pass the Py35 unittest temporarily.

* [Cherry-pick]Fix bug in convert_call because difference exists between python3 and python2. (#23966 )

* allow amp and recompute working together

Weight quantization support channel_wise_abs_max method to achieve higher accuracy (#23629) (#23937)

* Weight quantization support channel_wise_abs_max method to achieve higher accuracy

[Dy2stat] Add Test and Example Code for Different Access to ProgramTranslator and Fix Related Bug (#23958) (#23963)

Cherry pick of 23958

1. Rename Dygraph To Static Decorators to declarative
2. dygraph_to_static_func is still using in some training tests, I cannot delete it now.
3. Add some API docs

* [Cherry-pick Release/2.0] remove redundant visit func, test=develop (#23877)

* [Cherry-pick Release/2.0] [dy2static]Fix a bug of is_dygraph_api and move BasicApiTransformer to a separate file(#23923)

* Move BasicApiTransformer to a separate file. test=develop

* Fix a bug: A api in module is not a real dygraph api in dygraph_to_static. test=develop
Co-authored-by: NChen Weihang <chenweihang@baidu.com>

* Avoid logging.info be printed many times in dygraph_mode,test=develop

* Avoid logging.info be printed many times in dygraph_mode,test=develop

* OP clip, merge_lod_tensor, convert/elementwise error message enhancement. test=develop

* add addcmul, test=release/2.0

* enhance attention_lstm and param_attr error message

* test=develop fix api doc error

test=release/2.0

* Fix judge pslib transpiler cherry-pick (#23720)

fix compile problem on windows and some invalid argument check, cherry pick from #23831, test=release/2.0 (#23870)

Fix the compile problem on windows, cherry-pick from develop branch 

[cherry pick] fix sample code diff for OP(retinanet_detection_output, retinanet_target_assign), fix  default value for OP(sigmoid_focal_loss) (#23744). test=release/2.0-beta (#23858)

eagerly release cuda resources before cuda enviroment destroying
test=develop

Fix the support the float16 of scale op, add delete the raise case for fp16

* add approximation for gelu, test=develop

* add functional conv

* add test and doc for function convs, test=develop

* update ConvTransposeOp's InferShape and error message, test=develop

* add hsigmoid, row_conv in paddle.nn(functional and layer), test=develop

* fix hyperlinks in docstring

* correct reader device index, test=develop

* fix async executor scope var initialization, test=release/2.0

* Add variance computing layer for tensor

* Format input data types

* Remove unnecessary expand operation

* Some fixes in doc

add new tensor: max,min,t,eye,log1p; test=develop

Those ops add the kernel message enhancement, as follows
paddle.fluid.layers.elementwise_add	
paddle.fluid.layers.elementwise_div
paddle.fluid.layers.elementwise_floordiv
paddle.fluid.layers.elementwise_max	
paddle.fluid.layers.elementwise_min	
paddle.fluid.layers.elementwise_mod	
paddle.fluid.layers.elementwise_mul	
paddle.fluid.layers.elementwise_pow	
paddle.fluid.layers.elementwise_sub

* Fix elementwise compile error, test=develop

* split inplace_abn & batch_norm InferShape. test=develop

* fix type. test=develop

* fix message. test=develop

* fix ENFORCE. test=develop
Co-authored-by: Nzhaoyuchen <zhaoyuchen01@baidu.com>

* lod_tensor_to_array error message enhance. test=develop