Support batching for rbapinns

[GiovanniCanali]

Description

This PR fixes #588

In this PR, I have added the support for batching while using RBAPINN solver class.

Checklist

• Code follows the project’s Code Style Guidelines
• Tests have been added or updated
• Documentation has been updated if necessary
• Pull request is linked to an open issue

[dario-coscia]

I think the logic for batching weights is very clean, this actually solves the problem nicely!

I put some consideration on the code, I am wondering about future maintainability, I think we can clean it more

[GiovanniCanali]

I think the logic for batching weights is very clean, this actually solves the problem nicely!

I put some consideration on the code, I am wondering about future maintainability, I think we can clean it more

Hi @dario-coscia, thank you for your insightful review. Please, see the comments attached to your suggestions.

I understand the proposed RBAPINN implementation is a bit forced. However, I feel like the logic behind PINNs is way too complicated and messy at times, leading to weird artifacts when the training step and validation/test steps differ. Also, it is confusing to have both _optimization_cycle and optimization_cycle.

My suggestion is to discuss in the near future a reorganization of the whole logic behind the PINN-based solvers. Nothing would change under the user's perspective.