Fix onnxfs2_test regression for onnx 1.13

[BowenBao]

Root cause is new "Reduce ops" spec promotes axes from attributes to inputs, resulting in onnx test cases having more inputs than expected by these onnx-script tests. This PR updates these onnx-script test functions to match the signature of the updated onnx test cases.

However, since onnx test cases are not versioned, I feel tests like this that seemingly want to represent older opset onnx operator with function should not depend on it.

Fixes #249
Fixes #298

[codecov]

Codecov Report

Merging #307 (bcd1005) into main (42be235) will decrease coverage by 0.09%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #307      +/-   ##
==========================================
- Coverage   72.95%   72.86%   -0.10%     
==========================================
  Files          97       97              
  Lines        9436     9426      -10     
==========================================
- Hits         6884     6868      -16     
- Misses       2552     2558       +6     

Impacted Files	Coverage Δ
onnxscript/test/functions/onnxfns2_test.py	97.36% <ø> (-0.31%)	⬇️
onnxscript/test/models/onnxfns2.py	62.71% <100.00%> (-2.92%)	⬇️
onnxscript/type_annotation.py	85.71% <0.00%> (-11.43%)	⬇️
onnxscript/onnx_opset/_impl/opset16.py	71.42% <0.00%> (-9.53%)	⬇️
onnxscript/onnx_opset/_impl/opset13.py	88.07% <0.00%> (-7.34%)	⬇️
onnxscript/onnx_opset/_impl/opset12.py	49.39% <0.00%> (-3.62%)	⬇️
onnxscript/onnx_opset/_impl/opset11.py	55.29% <0.00%> (-3.53%)	⬇️
onnxscript/utils.py	56.25% <0.00%> (-3.13%)	⬇️
onnxscript/autocast.py	96.15% <0.00%> (-1.93%)	⬇️
onnxscript/onnx_opset/_impl/opset1.py	32.30% <0.00%> (-1.54%)	⬇️
... and 10 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[justinchuby]

Thanks a lot for fixing the ci! cc @gramalingam for comments around test usage in the PR description

[gramalingam]

Thanks, seems worth looking into a bit more. At the least, the definition of the Reduce functions does target opset18 (where the attribute was promoted to input), so onnxfns2.py should be importing opset18 instead of opset15. But I believe it should support the case where the axes is not specified or axes=None.

Adding @liqunfu .