Second round of cherry-pick (microsoft#6083)

* Fix PR microsoft#5550 reverted in microsoft#5911 (performance improvment for operator Transpose) (microsoft#5916)

* Improves implementation of transpose operator
* Fix issue mentioned in microsoft#5911
* adding unit test for function DoTransposeImpl

* Make operator TreeEnsemble 5x faster for batches of size 100.000 (microsoft#5965)

* improves processing time by 10
* extend coverage unit test coverage
* better implementation for the multi regression case
* better comment, keep parallelization by trees when not enough trees

* Initialize a structure in operator ReduceSum (microsoft#6005)

* fix initialisation issue

* Fuse MatMulIntegerToFloat only when scales are scalar (microsoft#6008)

MatMulIntegerToFloat fusion fuses per-row and per-column MatMulInteger, which is not supported by the MatMulIntegerToFloat kernel now. Limit the fusion to per-matrix only before we supporting the per-channel fully.

* Disable Python 3.9 for training Python packaging build. (microsoft#6012)

Disable Python 3.9 for training Python packaging build. Python 3.9 is not supported by the PyTorch dependency.

* Fix bugs for 1: Calibrator should check model inputs; 2: (microsoft#6017)

quantize_inupts forgot to use parameter initializer_use_weight_qtyp.

* Bump highlight.js from 10.2.1 to 10.4.1 in /nodejs

Bumps [highlight.js](https://github.com/highlightjs/highlight.js) from 10.2.1 to 10.4.1.
- [Release notes](https://github.com/highlightjs/highlight.js/releases)
- [Changelog](https://github.com/highlightjs/highlight.js/blob/master/CHANGES.md)
- [Commits](highlightjs/highlight.js@10.2.1...10.4.1)

Signed-off-by: dependabot[bot] <[email protected]>

* work around of the build break in mac (microsoft#6069)

* Fix the build break in macos release

* revert android change

* Bump up API version for 1.6 release (microsoft#6076)

* Update version to 1.6.0 (microsoft#6041)

* Update version to 1.6.0

* Add v 1.5.3 info

* Updating WindowsAI and ONNX version

Co-authored-by: Du Li <duli@OrtTrainingDev0.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>

* Rsevert "Fuse MatMulIntegerToFloat only when scales are scalar (microsoft#6008)"

This reverts commit beb950e.

Co-authored-by: Xavier Dupré <[email protected]>
Co-authored-by: Yufeng Li <[email protected]>
Co-authored-by: Edward Chen <[email protected]>
Co-authored-by: Zhang Lei <[email protected]>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Pranav Sharma <[email protected]>
Co-authored-by: Du Li <duli@OrtTrainingDev0.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>

Update Version to 1.5.3

Skip running onnx tests in python mac os pipeline (microsoft#5416)

orttraining_rc3.1 release

orttraining_rc3 release

fix quantization of EmbeddingLayerNorm (microsoft#5321)

fix quantization of EmbeddingLayerNorm (microsoft#5321)

For ORTTraining RC2 release to include hotfix microsoft#4639.

It will be used to create docker image for orttraining examples.

Onnxruntime 1.4.0 Release Tag

Cherry pick more commits to Rel-1.3.1 (microsoft#4231)

* Fix deprecated CentOS link for Linux CI pipeline (microsoft#4000)

* Fix Linux_CI_GPU_Dev

* centos6

* Fix crash reported in microsoft#4070. (microsoft#4091)

* Fix crash reported in microsoft#4070.

* Add newline to warning message

* Add comment for using cout instead of the logger

* fix optional input/outputs (microsoft#4229)

Co-authored-by: Ryan Lai <[email protected]>
Co-authored-by: Pranav Sharma <[email protected]>
Co-authored-by: Tracy Sharpe <[email protected]>