Pick many infer fix

Infer shape of pad_op pad_constant_like_op for version 1.4.0

[Cherry-pick] batch_norm and cos_sim infershape

test=develop

Merge pull request #16798 from phlrain/softmax_cross_support_high_rank

[Cherry pick] modified infer shape

[cherry-pick] infer shape: grid_sampler, kldiv_loss, spectral_norm, interpolate

[cherry-pick]polish doc for uniform_random and multi_box_head

[Cherry-pick] polish yolov3_loss annotation

add continuous value model op

refine roi align infer shape. test=release/1.4

* fix infershape in runtime

* fix infershape in runtime
test=develop

* fix infershape in runtime

* scatter_op bug fix, test=develop

* scatter_op bug fix, test=develop

* scatter_op bug fix, test=develop

separate runtime infershape

Fix merge_lod_tensor_op infer shape

Fix infer shape of split lod tensor op

fix teacher_student op infer

Fix sum infershape issue

Fix bpr loss

Fix conv_shift_op infershape

[cherry-pick] Fix load persistables in graph wrapper.

fix split_byref_op infer shape

Merge pull request #16746 from xjqbest/dataset_merge_develop

Merge pull request #16652 from xjqbest/dataset_merge_develop

test=develop

fix save and load bug, test=release/1.4

test=release/1.4

Enable setting thread number in Anakin Config

Fix code style

Reduce cmake file changes

test=release/1.4

imperative fix train mode

test=develop

[cherry-pick] fix yolo param

softmax cross entropy support high rank

move split filelist from trainer.py to fleet & fix error

fix dataset bug