删除Broadcast function中rank例化以及Elementwise调用，降低编译时间。
从develop分支中的#42645 PR修改而来，由于develop分支与release分支相差较大，无法实现cherry-pick，因此针对release2.3重新提交PR.
Broadcast中关于rank的例化会导致底层模板展开较多，造成reduce_sum_grad_kernel.cu.o文件体积过大，修改后可以降低.o体积及编译时间

* Support MaskedSelectGrad op with Kernel Primitive API

* Replace PADDLE_WITH_XPU2 with PADDLE_WITH_KP

* format softmax forward

* seperate include and macro to two if-else

* add gaussian random

* add full

* refine reduce

* refine code

* refine gaussian_random unittest

* add unittest for fill_any_like fill_constant

* Add function description for Kernel Primitive API
1. Set cumsum and sort share memory size = 1024
2.sort and cumsum api limitation : blockDim.x must be less than 512 (blockDim.x <= 512)

* [KP] Add elementwise add xpu, test=develop

* modify the File Permissions

* modify the copyright time

* modify code style

* modify code style

* Modify the implementation of BlockYReduce to fit more scenes

* fix

* fix

* Modified RandomKernel with Kernel Primitive API

* update pten.h to phi.h

* update

* update fullKernel

* Add Sort API for Kernel Primitive API

* update & -> ptr

* rename pten dir to phi

* rename namespace to phi

* rename infrt pten dir to phi

* resolve conflict

* rename pten to phi in cmake

* revert all infrt change

* change needed files

* fix infrt failed

* fix inference failed