Implement a class to perform inference with the MANN network

[GiulioRomualdi]

Before proceeding with the review of the PR I would like to discuss with you (@traversaro @S-Dafarra @paolo-viceconte) how to handle these two facts:

1. In 7172fa3 I implemented the Findonnxruntime.cmake. I tested only on linux. A possibility is to get onnxruntime with fetchcontent however:
   1. the installation procedure is system dependent. In linux we have to unzip a compressed folder and move the .so and h in the installation folder. In windows via nuget
   2. The installation does not ship any cmake machinery so I don't know how to handle the linking of the libraries in our framework (probably with add_library(<name> <type> IMPORTED [GLOBAL])
2. In 421da31 I also stored the model for the test. This is pretty heavy (6mb). we may think of storing the model somewhere perhaps in mann-pytorch and take with fetchcontent

[S-Dafarra]

1. the installation procedure is system dependent. In linux we have to unzip a compressed folder and move the .so and h in the installation folder. In windows via nuget

I guess we can stick to linux only for the time being

2. In 421da31 I also stored the model for the test. This is pretty heavy (6mb). we may think of storing the model somewhere perhaps in mann-pytorch and take with fetchcontent

That's not desirable. Can't you have a smaller model?

[traversaro]

Regarding the installation procedure, I would go for:

• Binary provided by onnxruntime for Linux/apt . I hoped it was static, but even shared will work fine. Discussing with @paolo-viceconte and @GiulioRomualdi I suggested to put this installation on blf, but probably it make more sense to have it on the superbuild.
• For conda-forge, just install onnxruntime from conda-forge . At the moment onnxruntime is not available on Windows, but for now we can just enable it on Linux/macOS .
• On vcpkg and homebrew we do not support this, as life is already too short to try to support homebrew and vcpkg.

[traversaro]

Binary provided by onnxruntime for Linux/apt . I hoped it was static, but even shared will work fine. Discussing with @paolo-viceconte and @GiulioRomualdi I suggested to put this installation on blf, but probably it make more sense to have it on the superbuild.

Done in robotology/robotology-superbuild#1387 .

[GiulioRomualdi]

The PR is finally ready to be reviewed. The model is now downloaded while compiling 🚀

[S-Dafarra]

Not really a comment for this PR, but we are getting multiple copies of this file. I wonder if we could move this logic within add_bipedal_test. Maybe something to automatically generate this file given a path. Anyhow, this is not blocking at all

[GiulioRomualdi]

I think I applied all your suggestions, so in theory, the PR could be merged. However, we need ami-iit/mann-pytorch#5 to get merged before this one (@paolo-viceconte)