[ONNX] Foward port new mx2onnx into master

[Zha0q1]

This pr forward ports all the new mx2onnx changes in v1.9.x (or v1.x as of 06/15/2020) and deprecates onnx2mx as discussed in #20063.

The expectation is that models generated with and exportable by MXNet v1.x should be continue to be able to export fine with MXNet 2.0.

Done:

1. forward-port mx2onnx
2. add operator tests (onnx updated to v1.8, onnxruntime updated to v1.7)
3. add one model test (resnet 50)
4. check mx2onnx wheel generation

TODO in future prs:

1. identify new np and npx operators that need new conversion functions
2. forward port gluoncv/gluonnlp model tests once they are compatible with MXNet 2.0

---

update on 06/30/2021
Since this pr 7152685 MXNet 2.0 is steering toward the new Gluon 2.0 interface and np and npx operators. This PR will also try to adapt the existing conversion functions to the new operators.

Supported Ops (WIP):

Np
absolute
zeros
ones
zeors_like
ones_like
full_like
arange
reshape(naive support)
concatenate
transpose
expand_dims
add
equal (missing scalar case)
not_equal (missing scalar case)
greater (missing scalar case)
less (missing scalar case)
greater_equal (missing scalar case)
less_equal (missing scalar case)
minimum
maximum
where
add (scalar)
subtract (scalar)
multiply (scalar)
divide (scalar)
power (scalar)


Npx
slice
arange_like
layer_norm
sequence_mask
reshape(naive support)
fully_connected
leaky_relu
activation
dropout
cast
softmax
box_nms
broadcast_greater
topk

[mxnet-bot]

Hey @Zha0q1 , Thanks for submitting the PR
All tests are already queued to run once. If tests fail, you can trigger one or more tests again with the following commands:

• To trigger all jobs: @mxnet-bot run ci [all]
• To trigger specific jobs: @mxnet-bot run ci [job1, job2]

---

CI supported jobs: [centos-gpu, windows-gpu, website, windows-cpu, clang, centos-cpu, miscellaneous, unix-gpu, unix-cpu, sanity, edge]

---

Note:
Only following 3 categories can trigger CI :PR Author, MXNet Committer, Jenkins Admin.
All CI tests must pass before the PR can be merged.

[waytrue17]

LGTM, thanks!

[leezu]

Thank you! Few small questions: Can python/mxnet/onnx/mx2onnx/LICENSE be deleted, as mx2onnx is part of mxnet and not a separate project?

[leezu]

GluonNLP v0.10 does not support MXNet 2. Have you tried with MXNet 2 / should the GluonNLP version be updated here?

[leezu]

I think "forward port gluoncv/gluonnlp model tests once they are compatible with MXNet 2.0" from the PR description addresses my question. So mx2onnx currently supports exporting a symbol obtained via guonnlp 0.10 to onnx, it's just that we need mxnet 1 to export that symbol from gluonnlp 0.10

[Zha0q1]

Yes that's correct, let me tweak the doc to note that explicitly thanks!

[leezu]

Dual licensed files should also be mentioned in the root LICENSE file. Further, as these files start with AL2 License text, why do you need to include them in license_header.py? If the license test fails, would it succeed if Line 17 (the line after end of AL2 text would be empty instead of #?

[Zha0q1]

the op_translation files were originally developed by Nvidia I believe so they were dual licensed https://github.com/apache/incubator-mxnet/blob/97a2b609042e8594c6b1ad655a74f7eef585cfd0/python/mxnet/onnx/mx2onnx/_op_translations/_op_translations_opset12.py. Let me also mention them in root LICENSE