[Inference] Make share_external_data supports bf16 and bool; fix while_op cache_inference_while_scope when using fleet_executor. (#56055)

* 1. make share_external_data supports bf16 and bool; 2. don't drop_kids when cache_inference_while_scope

* fix FLAGS_cache_inference_while_scope

* add unitest

* add unitest

* skip unitest when cudnn_version < 8100

* skip test share_external_data_bf16 when CUDA_ARCH < 80

* fix op translator reshape type

* new ir support vector type place transfer

* add test case

* fix new ir bn grad bug

* update

* update

* [Fluid] Migrate c_embedding to PHI

* fix

* add python_api

* fix ut

* migrate xpu kernel

* fix windows compile error

* support_mmha
* add_python_api
* add_api_doc
* fix_doc_error
* fix_infermeta
* add_infermeta
* add_bf16_cuda_check
* add_bf16_check
* fix_ci_windows
* fix_ci_windows_kernel_register
* fix_test_mmha
* add_cumoffsets
* remove_bias
* delete_mmha_reshape_input_output
* rename_delete_hfile
* remove_fluid

---------
Co-authored-by: Nyangjianfengo1 <yangjianfeng01@baidu.com>

* Improve GC for pipeline parallel

* Delete print

* fix bug of nop_op and sharding

---------
Co-authored-by: Nchenruibiao <chenruibiao@baidu.com>

* [Fluid] Migrate number_count to PHI

* fix out alloc

* fix ut (add python_api)

* onednn: rename macro to PADDLE_WITH_DNNL

* onednn: rename macro to CINN_WITH_DNNL

* move decayed_adagrad_op to phi

* fix bug

* Improve GC for pipeline parallel

* Delete print

* fix cudnn 8.7+ bug on cudnnConvolutionBiasActivationForward

* save_optimized_model_pass support tensorrt

* update

* update

* fix compile

* update

* fix ut timeout

* [clang-tidy] enable modernize-use-emplace

* Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into modernize_use_emplace

* move prune_gate_by_capacity to phi

* fix

* fix registe info

* remove useless codes

* new ir support print op

* fix gpu bug

* fix bug

* update

* remove layout to string

* remove usless header

* polish code

* fix bug

* posolis code

* use string as key for comm_context_manager

* remove device_id from comm_context

support stride

* rename BatchNormGradFunctor

* Update batch_norm_grad_kernel.cc

* Update batch_norm_grad_kernel.cu

* Update batch_norm_grad_kernel.cc

* fix

* Update batch_norm_grad_kernel.cc

* new ir support save combine

* update

* polish code

* update

* new ir support fluid op

* remove depulicate op

* fix ir exe test compile error

* fix compile bug

* update

* code format

* update

* update

* polish code

* Paddle-TRT support float64  in/out type, support fill_any_like_op in int64