Interleaved MHA for CPU path (#17138) (#17211)

* init qk attention

* qk: fake backward

* cpu selfatt and encdec; move tests to test_operator.py

* coding style

* fix lint

* use random seed in tests

* remove ut in test_operator_gpu.py

* coding style

* retrigger ci

Co-authored-by: Tao Lv <tao.a.lv@intel.com>

Author: 2 people