MAINT: Standardize deprecation calls and removed_in format

[romanlutz]

Summary

Two related cleanup tasks across the PyRIT codebase, both guided by the "Deprecations" section of .github/instructions/style-guide.instructions.md.

Task 1 — convert warnings.warn(..., DeprecationWarning, ...) → print_deprecation_message (15 sites across 9 files)

The style guide mandates pyrit.common.deprecation.print_deprecation_message (never warnings.warn directly) for deprecations of public API. Converted:

File	Sites
pyrit/datasets/seed_datasets/seed_dataset_provider.py	3
pyrit/prompt_target/common/prompt_chat_target.py	2
pyrit/prompt_target/azure_ml_chat_target.py	1
pyrit/prompt_target/openai/openai_image_target.py	3
pyrit/prompt_converter/azure_speech_audio_to_text_converter.py	2
pyrit/prompt_converter/azure_speech_text_to_audio_converter.py	1
pyrit/prompt_converter/persuasion_converter.py	1
pyrit/prompt_converter/variation_converter.py	1
pyrit/score/audio_transcript_scorer.py	1

Each conversion preserves the removed_in version the original message advertised. Unused import warnings lines were dropped after conversion.

Non-deprecation warnings.warn sites (UserWarning, FutureWarning, suppression context managers) were intentionally left alone.

Task 2 — strip the v prefix from removed_in= (3 sites)

Style-guide format is "0.15.0", not "v0.15.0".

• pyrit/setup/initializers/__init__.py
• pyrit/setup/initializers/pyrit_initializer.py
• pyrit/prompt_target/hugging_face/hugging_face_endpoint_target.py (also converted from warnings.warn per Task 1's spirit)

Task 3 (audit-driven) — normalize removed_in="2.0" → "0.16.0" (12 sites)

grep audit surfaced 12 print_deprecation_message calls in pyrit/output/ that all set removed_in="2.0" — a placeholder from the printer-refactor PR (#1732). Per the +2-minor-versions rule and the current 0.14.0.dev0 version, normalized all 12 to "0.16.0":

• pyrit/output/scorer/base.py (2)
• pyrit/output/scenario_result/pretty.py (1)
• pyrit/output/scenario_result/base.py (1)
• pyrit/output/attack_result/pretty.py (5)
• pyrit/output/attack_result/markdown.py (3)

One intentional outlier was left alone: pyrit/scenario/core/scenario_strategy.py uses removed_in="0.18.0" with an in-code comment explaining the extended window for Foundry callers.

Test updates (4 files)

A few test assertions checked unique phrasing from the original warnings.warn messages that the helper format doesn't reproduce verbatim. Updated:

• tests/unit/datasets/test_seed_dataset_provider.py — 2 patterns
• tests/unit/setup/test_pyrit_initializer.py — 2 patterns (dropped v from regex)
• tests/unit/prompt_target/target/test_image_target.py — 4 patterns (dropped v, adjusted quoting around style)
• tests/unit/prompt_target/target/test_normalize_async_integration.py — 1 pattern (message_normalizer is deprecated → message_normalizer + deprecated checked separately, to accommodate the helper's (param=...) suffix)
• tests/unit/prompt_converter/test_azure_speech_text_converter.py — 1 pattern (recognize_audio() does not support callable token providers → recognize_audio + deprecated)

Verification

• uv run pytest tests/unit -x -q → 8049 passed, 118 skipped
• uv run pre-commit run --all-files → all hooks pass (ruff format / ruff check / ty)
• grep "warnings\.warn.*DeprecationWarning" pyrit/ → zero hits outside pyrit/common/deprecation.py (the helper itself)
• grep 'removed_in="v' pyrit/ → zero hits
• grep 'removed_in="2\.0"' pyrit/ → zero hits