Enable custom model construction

[meilame-tayebjee]

This pull request introduces a new "custom model" architecture path to the documentation and examples, clarifies the two supported ways to build classifiers, and improves code style and reproducibility in the advanced training example. The most important changes are grouped below.

Documentation: Custom Architectures and Dual Path Explanation

• Added a comprehensive "Two Architecture Paths" section to architecture/overview.md, explaining both the standard (ModelConfig) and custom (from_model) approaches, including interface requirements and usage of the contrib package for reference implementations.
• Added a new tutorial (custom_model.md) detailing how to implement and use custom PyTorch models with torchTextClassifiers.from_model, including minimal and multi-task examples, interface requirements, and saving/loading.
• Updated the main README.md to clarify the two architecture paths and reference the new contrib-based multi-level classification example. [1] [2]

Tutorials and Learning Path Updates

• Added a "Custom Architectures" tile to the tutorials index with a new entry in the learning path diagram, linking to the new tutorial and outlining what users will learn. [1] [2] [3]

Example Code Improvements

• Refactored examples/advanced_training.py for improved code style, reproducibility, and clarity: removed unused imports, standardized environment variable usage, improved array construction, and cleaned up training configuration and model instantiation. [1] [2] [3] [4] [5] [6] [7] [8] [9] [10]