[DRAFT] FEAT Add Agentic Prompt-Injection Boundary Pairs dataset loader

[3nesdeniz]

Description

Closes #2174.

Adds a remote loader for the CC BY 4.0 Agentic Prompt-Injection Boundary Pairs dataset:

• 1,200 English records organized as 600 controlled benign/attack pairs;
• 50 scenario-isolated enterprise and agentic workflows;
• 12 prompt-injection and trust-boundary attack families;
• train, validation, test, or combined loading;
• attack-only default plus benign and full-pair modes;
• typed filters for label, split, attack family, and source context;
• pair, scenario, source, boundary, expected-action, and provenance metadata on each seed.

The loader maps each row to a literal SeedPrompt. Family filtering uses pair_family rather than attack_family, which preserves the benign side of a pair when users request the complete boundary surface.

Dataset: https://huggingface.co/datasets/3nesdeniz/agentic-prompt-injection-boundary-pairs

Source and release pipeline: https://github.com/3nesdeniz/agentic-prompt-injection-boundary-pairs

I am the dataset author and maintain the tagged source release. The dataset is synthetic, manually reviewed, and contains no customer or private data.

Tests and Documentation

• Added 15 focused loader tests covering defaults, all splits, benign-only loading, pair reconstruction, family and source-context filters, invalid enums, empty results, malformed rows, and split conflicts.
• Ran the full dataset unit-test package: 994 passed.
• Ran focused loader and metadata coverage tests: 80 passed.
• Ran all changed files through pre-commit, including Ruff, ty, notebook sanitation, documentation validation, private-key detection, and JSON checks.
• Performed a live Hugging Face sanity check:
  • test split attack mode: 120 seeds;
  • indirect-content-injection paired mode: 20 seeds / 10 complete pairs;
  • both labels preserved and prompt-group aliases reconstructed.
• Added the loader to dataset discovery, the loading guide and notebook, and the project bibliography.