[cherry-pick][hybrid performance] optim the grad fuse for pipeline mode by sorting the grad by dtype (#35070) (#35300)

[cherry-pick][hybrid performance] Grad fuse for gradient merge under pipeline mode (#35004) (#35299)

Co-authored-by: NWangXi <wangxi16@baidu.com>

* support ScaleTensor for scale npu kernel

* add more tests for adam npu

* fix compile

* fix unittest

* refine adam optimizer

* pass found_inf to adam

* add unittest

* fix bug

* refine unittest

* change unit test's directory

* disable unittest on cpu

* mv recv to head, test=develop

* cache core.ops, test=develop

* refine, test=develop

* opt dygraph python API, test=develop

* Fix unbind bug in manipulation.py

* enable npu alignment

* support flatten_params/grads

* support clip by global norm

* remove memset in coalesce_tensor_op

* fix npu kernel of sum op when input is one tensor

* add ut for flatten_param_grads+regularizer

* fix ut

* fix typo

This reverts commit b2afc8df.

* fix Undefined variables

* fix Undefined variables

* fuse L2Decay and momentum when param.regularizer is set

* add unittest

* refine

* refine _create_regularization_of_grad of momentum

* improve append_optimizer_op

* update, test=develop

* add use_global_beta_pow

* add use_global_beta_pow

* update npu kernel

* update python api

* refine code

* add ut for use_global_beta_pow

* fix npu kernel

* add ut for api

* add ut for exception

* add ut for save/load

* add input EpsilonTensor for adam

* update python api

* add unit test

* add npu test

* add more ut

solve hccl communicate conflict (#32447)

fix test sync_with_cpp (#32212)

* update, test=develop

* update, test=develop