FIX: Use sequence=0 for both pieces in multimodal dataset loaders

[romanlutz]

Description

Four multimodal remote dataset loaders were assigning sequence=0 to one piece (image or text) and sequence=1 to the other while sharing the same prompt_group_id. Per SeedPrompt.sequence (pyrit/models/seeds/seed_prompt.py:43-44), prompts are only grouped into a single multimodal user message when they share both prompt_group_id and sequence. With mismatched sequences, the image and text were being delivered as two separate turns rather than as a single multimodal message, which defeats the purpose of these datasets (the model is supposed to reason over image + text together).

This PR brings the four affected loaders in line with the correct pattern already used by harmbench_multimodal_dataset.py and the recently added msts_dataset.py: both pieces share prompt_group_id and sequence=0.

Loader changes (pyrit/datasets/seed_datasets/remote/):

• vlguard_dataset.py - image sequence=1 -> 0.
• vlsu_multimodal_dataset.py - image sequence=1 -> 0.
• visual_leak_bench_dataset.py - text sequence=1 -> 0. Reworded class and fetch_dataset_async docstrings that described the old behavior.
• comic_jailbreak_dataset.py - text sequence=1 -> 0. Reworded fetch_dataset_async and _build_seed_group docstrings. The SeedObjective in the group is unchanged - only the image+text pair needs to share sequence=0.

Tests and Documentation

Updated the four corresponding unit tests under tests/unit/datasets/ to assert the new shared sequence == 0 for both pieces (one assertion change per file).

• uv run ruff format pyrit tests - clean
• uv run ruff check pyrit tests - clean
• uv run -m ty check pyrit/datasets/seed_datasets/remote - clean
• uv run pytest tests/unit/datasets -q - 422 passed
• pre-commit hooks passed on commit

No JupyText/doc changes needed (no docs reference these sequence numbers).