Fix the bug of slice op and optimize the code style of generate_proposals_v2 op for kunlun. *test=kunlun (#43380)

* change tensor name

* fix unittest bugs

* make all phi kernels to 2(host/device) static libraries directly

* fix calling kernel_declare

* fix compile

* fix cpu compile

* fix rocm compile

* fix xpu compile

* fix xpu kp compile

* fix inference compile

* add new field for event node

* fix

* fix bug

* fix bug

* fix clang

* fix clang format

* fix code format

* add unary ops

* move to activation_ops

* Refactor DeviceContextPool

* Adjust header file order

* Re-implemented check_finite_and_unscale_op  with newly added xdnn api
*test=kunlun

* Re-implemented check_finite_and_unscale_op  with newly added xdnn api

*test=kunlun

* fix some matmul that X and Y both other ops outputs, do not dequantize the Y.

* fix CI format

* fix according to review

* add nproc_per_node for DistributedFusedLamb

* fix nproc_per_node communicator bug

* fix ring_id = 1 init bug

* fix ci

* fix test_parallel_executor_mnist.py

* fix sign-compare warning

* fix sign-compare 2

Co-authored-by: NLiu Yiqun <liuyiqun01@baidu.com>

* fix pylayer forward output code

* refine

* move_group_norm

* move group norm backward

* fix code format

* modify code according comment

* support matmulv2 in multihead

* Polish codes and memory usage for fused_gate_attention.

* Fix wrong reduce_dims in fused_gate_attention when computing gradient of nonbatched_bias.