Refactor and optimize the dataset package interfacing for Go idioms (reference: PyTorch)

[gitctrlx]

Background

The current dataset package in this framework could be streamlined for better usability and maintainability. PyTorch offers a robust and flexible dataset/dataloader design which is widely recognized in the ML community. However, Go's language features and developer habits are significantly different from Python.

Proposal

• Analyze PyTorch's Dataset and DataLoader implementations as reference.
• Redesign and refactor the dataset package to offer interfaces, conventions, and usage more idiomatic for Go users.
• Provide concise examples demonstrating both basic and advanced usage.
• Ensure loose coupling, testability, and extensibility.
• Document design decisions and migration guidance from old APIs.

Goals

• Deliver an ergonomic and Go-centric dataset interface while retaining the power and flexibility seen in PyTorch.
• Reduce friction for Go ML developers transitioning from PyTorch or other Python-based frameworks.