* Add `paddle.nn.loss.MSELoss`

test=develop

* Move to `nn/layer/loss.py`

test=develop

* Add test

test=develop

* Fix dygraph

test=develop

* Increase numel in test

test=develop

* Add test for input with more dimensions

test=develop

* test=develop
Co-authored-by: Nwuxing03 <wuxing03@baidu.com>

Fix bug in convert_call because difference exists between python3 and python2. test=develop (#23966)

restrict block num of layer_norm_grad cuda kernel to 128, test=develop

* Support LoDTEnsorArray in fetch op

test=develop

* Support LoDTensorArray in fetch

test=develop

* Support LoDTensorArray in fetch

test=develop

* Support LoDTensorArray in fetch

test=develop

* Support LoDTensorArray in fetch

test=develop

* Support LoDTensorArray in fetch

test=develop

* Support LoDTensorArray in fetch

test=develop

* Support LoDTensorArray in fetch

test=develop

* Support LoDTensorArray in fetch

test=develop

* Support LoDTensorArray in fetch

test=develop

* [Dy2Stat]Support recursive call.

* Remove Redundant decorator to pass the Py35 unittest temporarily.

* add gradient check to reduce ops

* add skip gradient check to reduce ops test=develop

* modify stack api test=develop

* add bmm op and move serval ops from fluid.layers to tensor test=develop

* add paddle.tensor
test=develop

* polish gpu where codes
test=develop

* polish test code
test=develop

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

To prepare for publishing APIs, I added tests for that we can access dy2stat through:

@fluid.dygraph.declarative
@fluid.dygraph.jit.declarative
fluid.dygraph.ProgramTranslator()
fluid.dygraph.dygraph_to_static.ProgramTranslator()
fluid.dygraph.dygraph_to_static.program_translator.ProgramTranslator()

It surprised me that we had bugs on those different usages. I have fixed them.

I also added example codes for these new APIs

This PR also pulls my current PR https://github.com/PaddlePaddle/Paddle/pull/23880, so the PR history is long. For reviewer information, you could review this PR after https://github.com/PaddlePaddle/Paddle/pull/23880 is merged

 fix compare and logical ops cond/out parameter doc bug，correct the dtype message 

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

* enhance error message test=develop

* fix CI test of 3 op test=develop

* fix bug caused by the diff of long type in py2 and py3, test=develop

* fix error message of l2_normalize, matmul, mean, etc. 
* add the test case for those ops

* fix trt fc fuse test, test=develop

* fix trt_transpose_flatten_concat shape, test=develop

* add addcmul, test=develop

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

As the title

* add nll_loss op test=develop

* add paddle.std api test=develop

* update  test=develop

* fix example code format test=develop

* allow amp and recompute working together

API/OP (center_loss, fluid.one_hot, prroi_pool, roi_pool, ctc_greed_decoder) error message enhancement (#23794)

* error message enchanced, test=develop

* error message enchanced for APIs, test=develop

* error message enhanced for roi_pool, test=develop

* update added code, test=develop

* update fluid/input.py, test=develop

* update fluid/input.py ,test=develop

* fix code style, test=develop

* fix lower converage, test=develop

* error message enhanced for roi pool, test=develop

As the title.

* fix bug in reduce op test=develop

* fix bug in reduce op test=develop

* error message enhancement create_parameter create_global_var, test=develop

* use six.PY2 to use long, test=develop

* fix input dtype for create_parameter, test=develop

* enhance the error message of box_coder, test=develop

* enhance error message in box_decoder_and_assign

* enhance error message of collect_fpn_proposals

* enhance errror message of distribute_fpn_proposals

* enhance error message of multiclass_nms, test=develop

* add nn.interpolate, test=develop

* fix interpolate typo, test=develop

* formate code, test=develop

* fix unitest, test=develop

* add test layers, test=develop

* add test layers, test=develop

* extract common function, test=develop

* reduce the threads for cuda10, test=develop

* update unitest, test=develop

* polish unitest, test=develop

* add dygraph test, test=develop

* format description, test=develop

* add 5D input check, test=develop

* fix doc, test=develop

test=develop

* enhance error message test=develop

* enhance error message test=develop

* change to INOUT_CHECK  test=develop

* add cudnn compatiable rnn cell api for dygraph

* update sample code

* update some typos

* fix specify name in param_attr problem

* add pre-commit check

* remove duplicate import, test=develop

* add unittest coverage, test=develop

* make code more tight, test=develop

* cudnn_compatibale -> use_cudnn_impl, test=develop

* change api name, test=develop

* test=develop fix api doc error

Fix the scale fp16 dtype support for scale op, and delete the test case for fp16 case

* `l1_norm` error message enhancement

test=develop

* `norm` error message enhancement

test=develop

* `squared_l2_norm` error message enhancement

test=develop

* `squared_l2_distance` error message enhancement

test=develop

* `conv_shift_op` error message enhancement

test=develop

* `sample_logits_op` error message enhancement

test=develop

* Fix format

test=develop

* use `PADDLE_ENFORCE_EQ` for '||' conditional

test=develop

* Fix string format

test=develop

* Add test for `sample_logits`

test=develop

* Add error message

test=develop

* Use ordered dict for outputs

test=develop

* Hard code fetch list so that order is deterministic

test=develop

* Fix license header year

test=develop