MAINT: migrate AdversarialBenchmark off deprecated PromptChatTarget

[romanlutz]

Description

pyrit/scenario/scenarios/benchmark/adversarial.py still typed its public adversarial_models parameter as list[PromptChatTarget]. PromptChatTarget was deprecated in #1678 with removal scheduled for v0.16.0, so without this migration the public AdversarialBenchmark(adversarial_models=...) API will break at import time once v0.16 ships.

Why it slipped

Merge race between #1662 (Adversarial Benchmark Scenario, merged 2026-05-07) and the surrounding deprecation work:

• FEAT Migrate consumers to TargetConfiguration capability checks #1645 (2026-04-30) migrated every other consumer to CHAT_TARGET_REQUIREMENTS.validate(target=...).
• FEAT: Adversarial Benchmark Scenario #1662 (2026-05-07) introduced this new file typed against the old API.
• FEAT deprecate prompt chat #1678 (2026-05-08) added the deprecation warning without rebasing onto the just-merged file.

No follow-up PR addresses it; this is the only remaining PromptChatTarget reference in pyrit/ outside the deprecated shim itself.

Changes

• Type adversarial_models as list[PromptTarget].
• Validate each supplied target against CHAT_TARGET_REQUIREMENTS in __init__ so callers get a clear, early error if a non-chat target sneaks in. Per-technique stricter requirements (e.g. SYSTEM_PROMPT for RedTeamingAttack) continue to be enforced at factory.create() time, unchanged.
• Update docstrings and error messages to describe the capability requirements rather than the deprecated class.
• Update test mocks to stamp a real TargetConfiguration (multi-turn, multi-message-pieces, system-prompt, editable-history — matching what PromptChatTarget._DEFAULT_CONFIGURATION advertised) so the existing test suite still exercises the techniques that demand these capabilities.
• Add a unit test asserting that an adversarial_models entry missing chat-target capabilities is rejected at construction.
• Sweep the two doc/scanner/benchmark.{py,ipynb} comments that referenced "PromptChatTargets".

Tests and Documentation

• uv run pytest tests/unit/scenario/test_adversarial.py — 34 passed
• uv run pytest tests/unit/scenario/ — 579 passed
• uv run ruff format --check + ruff check + ty (type check) — pass via pre-commit
• Doc references in doc/scanner/benchmark.{py,ipynb} updated to the new wording.

[hannahwestra25]

thanks ! was this caught by our integration tests ? I could be missing something but it seems like we don't have any integration tests running the doc/scanner notebooks

[romanlutz]

@hannahwestra25 copilot found this while doing a general assessment of the repo. It said something like it's a result of your PR (which removed it) merging around the same time as the adversarial benchmark PR got merged. I don't think the integration test was run because those happen outside of our PR gates.