* Add Multihead matmul fuse pass (#20167)

* Add multihead fuse pass for ernie opt

* Refine softmax

test=develop

* Refine cuda kernel

* Refine cuda version

* Refine cmake

test=develop

* refine header file

* refine test case and pass
* refine comments

* Delete useless code.

test=develop
Signed-off-by: Nzhaoyuchen <zhaoyuchen01@baidu.com>