Dataload

[JoKircher]

Feature: Introduce Data Loading Infrastructure

Summary

This PR merges the dataload branch into main, introducing core data loading and batching utilities for GraphNetCore.jl. These additions provide a standardized way to ingest graph data, handle feature matrices, and prepare mini-batches for downstream graph neural network training.

Key Changes

• Graph DataLoader: Implemented a generic data loader structure for handling collections of graphs.
• Batching Logic: Added efficient batching mechanisms to combine multiple disconnected graphs into a single large disjoint graph (e.g., concatenating adjacency matrices and node features).
• Dataset Support: Added utilities for downloading and parsing standard graph datasets (e.g., [Cora / CiteSeer / custom formats]).
• Preprocessing Pipeline: Introduced helper functions for normalizing node features and adding self-loops.

Motivation

Previously, users had to manually manage graph batching and feature alignment. This update abstracts the boilerplate data pipeline, bringing our API closer to standard GNN workflows and making it significantly easier to train models on large datasets.

Testing & Validation

• Added unit tests for the batching and unbatching logic.
• Verified data integrity (node indices, edge indexing) post-batching.
• All CI/CD pipelines and existing tests pass successfully.

Checklist

• Documentation updated to reflect new data loading structs and functions.
• Code follows the GraphNetCore.jl style guidelines.
• Exported new functions in the main module file.