Add Similarity Net as unit test. During the unit test, we found three problems:

1. The run_program_op has memory optimization error when running dy2stat net multiple times.
2. The support for SelectedRows can cause problem in dy2stat.
3. The return grammar has problem.

This PR fixes the 1. problem but modify codes for the 2. 3. problems to make PR smaller. I will fix those two problems in the next PR(s)

* Add the imperative quantization aware training.
*  This is the python part of Imperative QAT. test=develop

* fix softmax_with_cross_entropy cuda kernel overflow bug, test=develop

* replace old macro & for condition, test=develop

* polish details, test=develop

* add explicit specialization

* add skiplayernorm vector load if available

* test=develop

* fix concat shape error
test=develop

* add retry for prefetch

* Revert "Fix/sync barrier (#25016)"

This reverts commit be6a315f.

* reopen dist UT, test=develop

* remove fl UT, test=develop

* fix bn & in in dy, test=develop

* update instance_norm,test=develop

* fix bugs,test=develop

* add more case in unittest,test=develop

* fix,test=develop

* fix,test=develop

* add mish op. test=develop

* Add `matrix_nms_op`

test=develop

* Make ci happy

test=develop

* Exit early when no detection

test=develop

* Fix license year

test=develop

* Output index as well

test=develop

* Match nms2 lod behavior and add `return_index` flag

test=develop

* Make CI happy

test=develop

* Fix wording

test=develop

* test=develop, fix_embedding

* fix beam_search infershape, test=develop

* fix beam search op unittest, test=develop

* Refactor of conv fp32 oneDNN operator
test=develop

* Formatting fix
test=develop

* Return Enforces
test=develop

* GetWeights improvements
test=develop

* don't re-generate header file if content doesn't change. test=develop

* add copy_if_different function. test=develop

* Start implementing int8 eltwise add

test=develop

* - Fix to Michal PR

* - Fix

test=develop

* - Lint fixes

test=develop

* - Added checking if elementwise_mul can be used

test=develop

* - Added attribs to skip_attrs_set

test=develop

* - Improved broadcasting

test=develop

- fixes to compilation

- fix

- fix

- Lint fixes

test=develop

* - removed redundant condition

test=develop
Co-authored-by: NMichal Gallus <michal.gallus@intel.com>

* fix bug of prelu when rank not equal 4, test=develop

* fix prelu inference, test=develop

* fix api, test=develop

* fix shape when mode is chennel, test=develop

* remove debug code, test=develop

* add unittest, test=develop

* blockReduce opt

* launch threads align to warpSize

* reduce unnecessary shared memory for broadcast reduced value

* vectorize SoftmaxKernelWithEltadd

* add fp16 constrain

* test=develop

* test=develop

* test=develop

* fix bug

* test=develop

* test=develop

* fix sync barrier with barrier monitor, test=develop

Add the support the 5d,6d tensor support for the reduce ops;
Add the same time, the compile time, it was 22 minutes, it was 21 minutes after fixed.

fixes the CUDAPlace info in the Print op

* Support LoDTensorArray in reverse_op test=develop

* polish en doc and unittest code test=develop

* refine sample code test=develop

* add example of LoDTensorArray test=develop

* fix typo test=develop

* refine isfinite, test=develop

* use namespace std of isfinite, test=develop, test=win_gpu