Test inference performance on robot hardware

[texhnolyze]

Summary

While we have evaluated the latest model on the actual Wolfgan-OP platform, the model inference was running on one of our servers and published JointCommands over the network via ros messages.

To evaluate the performance criteria of the distilled model we should convert it to vulkan shaders for our Ryzen APU on the robots with tvm