related codes in Paddle core.

to create a gradient machine for inference using merged model with parameters
which is genearted by `paddle merge_model`.

* Should always use `dynload::` for cuda function.
* Fix cublas.h without DSO load.

Fix #3386

They are tested in Python

* Give which parameter, which element are wrong. And what max_diff is.

* No POD instance is forbidden in global scope.

See
https://google.github.io/styleguide/cppguide.html#Static_and_Global_Variables

* Make `PADDLE_ENFORCE_EQ` supports custom class, like DDim

* merge InferShapeContext and OperatorContext

* OperatorBase& instead of OperatorBase*

* add lodtensor

* add reshape of lod

* add details

* rename Elements/Levels

* size_t and vector reserve

* add details

* add const& std::shared_ptr

* add lod_tensor_impl.h

* remove a shared_ptr

* init grad op checker

* can run

* add GradeChecker class

* use get_numeric_gradient

* refine code

* add softmax and cross entropy auto grad test

* use close to judge op_grad and numeric_grad

* add cpu and gpu compare

* add comments

* add support_gpu

* fix allclose

* fix name error and symplify code

* optimize gradient checker

* add test_cross_entropy_op

* update gradient_checker.py

* optimize code

* use random.uniform instead of random.random

* fix type bug

* optimize check_grad

* put SupportGPU into OperatorBase

* typo