Change interface for nak_torch

[dannys4]

This pull request introduces significant refactoring and modernization to the example scripts and algorithm API usage in the project. The main changes involve updating the usage of particle-based inference algorithms to use new class-based APIs, centralizing the execution logic through a new nak function, and improving reproducibility and clarity in the examples. There are also minor improvements and cleanups in type usage and plotting.

Refactor and API Modernization

• Updated all example scripts (examples/gaussian.py, examples/himmelblau.py) to use the new class-based APIs for algorithms such as MSIP, SVGD, GradALDI, GradFreeALDI, CBS, and EKS, replacing previous function-based calls. The new nak function is now used as a central entry point for running these algorithms, leading to more consistent and maintainable code. [1] [2] [3] [4]

• Now use explicit random number generators (torch.Generator) for reproducibility, and initialization of particles and random seeds is more explicit and consistent. [1] [2]

Type and Interface Cleanups

• Several type hints and batch evaluation interfaces are improved, with more consistent use of batched log-density and gradient evaluators. [1] [2]

• Minor type and import cleanups, such as using DeviceLike from the project’s types module instead of redefining it in examples/pyro_tools.py.

Miscellaneous

• Minor changes to plotting and function signatures to support the new interfaces and improve code clarity, such as making auxiliary arguments explicit in functions and cleaning up unused code. [1] [2] [3]