Leveraging framework specific math helpers

[RogerChern]

Hi @wkcn , really nice work.
I'd like to ask is it possible for Mobula to leverage existing math helpers in deep learning frameworks like ATen in Pytorch and Mshadow in MXNet?
Writing everything from vanilla C++ is prohibitively cumbersome and thus prevent the adoption of Mobula in real practice.

[wkcn]

@RogerChern
Thanks for your support!
Currently, MobulaOP doesn't support math helpers.
I'm going to refactor MobulaOP with C++, and add the support of ONNX operators, which play the roles of math helpers : )