* add fp16 backward support
test=develop

* add sum_op fp16 test

* disable test_dist_save_load
test=develop

* add check_grad for sum

* add unit test for softmax_grad fp16
test=develop

* add scale_op unit test

* add mul_grad_op unit test for fp16

* add cross_entropy_grad and eman_grad unit test for fp16
test=develop

* fix cross_entropy unit test

* add pool2d fp16 unit test

* refine conv2d fp16 unit test
test=develop

* refine activation unit test
test=develop

* fix ci
test=develop

* follow zhihong's comment, copy from https://github.com/PaddlePaddle/Paddle/pull/12796
test=develop

This reverts commit bf3c3496.

* "cherry picked operators changes"

* "remove duplicated code"

* "add constant setter"

* "add get expected kernel"

* "fix ci"

* "add fill constant"

* "fix double type error"

* "fix ci"

* add fp16 mul op support

* small fix

* fix bug

* small fix

* fix PADDLE_WITH_CUDA compiling issue

* reorg code

* test for pybind

* treate as float16 as uint16_t in pybind

* bind np.float16 to paddle float16

* small fix

* clean code

* remove redundancy

* fix mul_op test

* address comments

* small fix

* add is_float16_supported func

There are mainly following fixes:

- take `DeviceContext` as the template parameter of math functors and OpKernel instead of `Place`
- remove `eigen_device` interface in base class  `DeviceContext`
- remove `GetEigenDevice` interface in `ExecutionContext` and base class `DeviceContext`
- remove unused `platform::EigenDeviceConverter`
- rename `REGISTER_OP_GPU_KERNEL` to `REGISTER_OP_CUDA_KERNEL`
- rename `USE_GPU_ONLY_OP` to `USE_CUDA_ONLY_OP`

Make implement an operator less noisy.

* Implement InferShape and register them, give a stub Kernel method
  by LOG(INFO)

* add NDEBUG switch to PADDLE_ENFORCE

* Free will be added soon