node converters for handling argmax and argmin onnx operators

[ajstarna]

• This PR contains the node_conversion code and tests for the ArgMax and ArgMin onnx operators (together known as arg extrema).
• the extrema functions as provided by torch do not handle the select_last_index attribute of the onnx operators, which was added with opset_version 12.

Notes to Reviewer:

• I added an is_cuda_available() check before running the cpu_cuda check function in the tests. Not sure if there is another way you would rather handle being able to run the tests on cpu only?
• using pdm for development, I was only able to install the package if I modified the required python version to be >=3.8 in the pyproject.toml file. Are we sure that all dependencies work under python 3.6 or 3.7?
• I was only able to run the tests by downgrading my onnx version from 1.16 to 1.15, or else I would get an error about supported IR version.

Thanks!

[ajstarna]

Hi @ivkalgin @senysenyseny16 - Any thoughts or feedback on this? 😃 Thanks!