FEAT: Add JailBreakV-28K dataset loader

[diamond8658]

FEAT: Add JailBreakV-28K remote dataset loader

Description

This PR introduces a new remote dataset loader for the JailBreakV-28K (V0.2) benchmark. It enables the seamless ingestion of over 28,000 jailbreak prompts directly from the source CSV hosted on the SaFo-Lab repository. This resolves issue #1007

Key Changes:

• Asynchronous Ingestion: Implements _JailBreakV28KDataset inheriting from _RemoteDatasetLoader for high-performance data fetching.
• CSV Version 0.2 Support: Specifically targets the versioned CSV release to ensure reproducibility and stability.
• Policy Mapping: Maps raw SaFo-Lab codes (P1-P5) to human-readable harm categories (e.g., Somatic Safety, Public Interest) to maintain consistency with existing PyRIT datasets.
• Defensive Filtering: Includes a safety check to skip prompts containing Jinja2 syntax ({{, {%) to prevent unintended orchestrator execution.
• Metadata Enrichment: Preserves raw policy codes and categories within the SeedPrompt metadata for granular downstream analysis.

---

Tests and Documentation

Unit Testing:

• Added comprehensive unit tests in tests/unit/datasets/test_jailbreakv_28k_dataset.py.
• Verified 100% test coverage for the new loader class.
• Handled edge cases including:
  • Malformed/missing CSV columns.
  • Jinja2 safety filtering logic.
  • Unknown policy code fallbacks (defaulting to "Unknown Policy").
  • Empty remote responses (raising ValueError).

Linting:

• Verified that the code passes all flake8 and ruff checks.
• Follows the Microsoft pyrit coding standards and naming conventions.

JupyText & Documentation:

• This PR implements a backend data loader and does not modify existing .ipynb or .md documentation files.
• No JupyText synchronization was required as no new notebook-based tutorials were introduced in this PR.
• Verified the loader's output format is compatible with the SeedDataset model used throughout the library's existing documentation and orchestrators.