Implement Magnetostatics in TorchPME

[E-Rum]

This branch is linked to #54 and introduces the initial implementation of magnetostatics in the existing TorchPME library. The commit provides a basic implementation for direct potential calculations along with a simple test case.

With this implementation, it’s a good time to start discussing how to structure the current magnetostatic functionality and align it with the broader scope of the library. Notable changes to the main library include:
1. Dipole Potential Components:
Unlike charge-based potentials (e.g., Coulomb), dipole potentials consist of two components: a scalar and a tensor part. Currently, these components are explicitly returned separately and later combined in the calculator. While this approach works for now, we should evaluate whether it remains viable as we transition to more complex Ewald-based calculators.
2. Calculator Structure for Dipoles:
The dipole calculator significantly differs from charge-based calculators, potentially necessitating a new base calculator tailored for magnetostatics. For now, the implementation resides in the calculators folder, but it may need restructuring.
Key differences include:
• Passing dipoles instead of charges.
• Utilizing neighbor vectors instead of neighbor distances.
• Returning potentials in a vectorial form, which appears more convenient from an ML perspective.

---

📚 Documentation preview 📚: https://torch-pme--133.org.readthedocs.build/en/133/