Package lie derivative code into neuralyzer

[dgcnz]

Issue:

• Currently the lie derivative code (https://github.com/ngruver/lie-deriv) requires you to clone submodules timm (pytorch image models) and stylegan3, which is completely unnecessary and clunky
• Furthermore, there is no installable package with no package lock and versioning

Tasks

• [ ] Understand basic requirements of lie-deriv function (inputs? outputs? usage?) [coordinate with @SebastianUrielArias]
• [ ] Copy and adapt relevant code from ngruver/lie-deriv into dgcnz/neuralyzer
• [ ] Document usage

[Nesta-gitU]

hey I like that you are making the library and it might also be useful in the future, but could you make me a collaborator on it, then I can also contribute the hessian spectra full code for example later.

[dgcnz]

@Nesta-gitU invited, see https://github.com/dgcnz/neuralyze