Differentiable Cheetah

[jank324]

🚀 Feature Request

Make sure that all or as many as possible computations in Cheetah are fully differentiable using PyTorch's built-in differentiation features.

Motivation

• Differentiation would make solving many problems super simple. One could then, for example use gradient descent to find magnet settings for the optimal beam.
• Differentiation would also be very simple to add, as gradient tracking is already done by PyTorch. All we would need to do is to make sure they track properly through all of Cheetah, i.e. we have low-hanging fruit here.

[jank324]

While probably not strictly needed, I think it makes sense to solve #10 before this issue.

[jank324]

This was done in #70, but apparently the issue was not linked to the PR and therefore not closed automatically.