* modify custom op dependent from paddle_framework to paddle_custom_op (#31195)

* [Custom Op] Remove unsupport dtypes (#31232)

* remove remove_unsupport_dtype

* remove remove_unsupport_dtype

* remove test dtype

* add more include

* change dtype.h's enum as enum class to avoid conflict with inference lib

* make enum as enum class

* remove additional test

* merge develop

* polish code

* [Custom OP] Support stream set on Custom Op (#31257)

* [Custom OP] change the user header file format, test=develop (#31274)

* [Custom OP]add PD_THROW and PD_CHECK for User Error message (#31253)

* [Custom OP]add PD_THROW and PD_CHECK for User error message

* PD_THROW and PD_CHECK, fix comment

* fix Windows error message

* fix Windows error message

* fix CI

* [Custom OP]add MSVC compile check on Windows (#31265)

* fix test_check_abi
Co-authored-by: NZhou Wei <52485244+zhouwei25@users.noreply.github.com>
Co-authored-by: NJiabin Yang <marsyang199376@gmail.com>
Co-authored-by: N石晓伟 <39303645+Shixiaowei02@users.noreply.github.com>
Co-authored-by: Nzhouwei25 <zhouwei25@baidu.com>

[Cherry-pick] The Second part of new custom op extension in 2.0.1

* test=develop, fix a bug

* test=develop, remove the macro of PADDLE_DLL_INFERENCE

* add SetCommandLineOption

* add the print_FLAGS function

* remove the test demo

* modify the location of macro

* add the 'WITH_STATIC_LIB' option on windows

* modify the macro of PD_INFER_DECL

* modify the the fuction name

* modify the unittest

* modify the code style

(1) use channel instead of vector/BlockingQueue in Dataset，to keep same with existing implementation, and make code more readable and flexible (dataset single output channel or multi output channel). one previous memory out of limit problem is cause by not release memory after training.
(2) add Record because MultiSlotType costs too much memory (80B)，fix memory out of limit problem.
(3) add Channel, Archive in paddle/fluid/framework
(4) change dataset from shared_ptr to unique_ptr in pybind
(5) move create/destroy readers from trainer to dataset
(6) move shuffle from datafeed to dataset. dataset holds memory, datafeed is only for load data and feed data to network.
(7) fix thread num bug of Dataset when filelist size < thread num
(8) support set_queue_num in InMemoryDataset

test=develop

add python inference api

fix cmake again
test=develop

Add the demo of subgraph splitter

* Update tensor_util.h

* Update with moved TensorDesc

* Fix tensur_utils.cu

* Update

* Update

* Update

* Update

* Make tensor_util.cu a symbolic link