promptinj: split AltayDuel cases into schema files; name-only adapter credit

[emmanuelgjr]

Summary

1. AltayDuel cases now follow the dataset schema

tr_altayduel_testcases.json (5 cases, contributed by @fevziegeyurtsevenler in #8) was a single aggregate document with its own field names. None of its cases matched promptinj_testcase.schema.json, and the dataset README asks for one file per case. They are now tr_altayduel/cases/TC-0451.json to TC-0455.json:

• id became testcase_id (TC-0451 to TC-0455, the next free IDs); the original TR-ALTAYDUEL-00N is kept in source_id.
• category uses the schema's values; dsgai_mapping is a list; expected_behavior_secure/_vulnerable and severity_if_bypassed are renamed expected_secure/expected_vulnerable and severity.
• The prompts are byte-identical, including the intentional Cyrillic homoglyphs in TC-0454. Glosses, expectations, prerequisites, severities, sources and notes are also unchanged.
• tr_altayduel/README.md records the contributor, the CC BY 4.0 source attribution and these format changes. The dataset README's suite table points at the new folder.

2. Name-only adapter credit

In tr_altaysec_turkish_llm_injection/ (contributed by @3nesdeniz in #11), provenance.adapted_by in all 300 cases and the README's "Adapted by" row now read Enes Deniz instead of Enes Deniz / AltaySec. This follows the repository rule that contributors are credited by name only (as in @3nesdeniz's later #46, which credits "Enes Deniz").

Deliberately left unchanged:

• Source creator and copyright (Fevzi Ege Yurtsevenler / AltaySec). The source dataset's LICENSE gives that copyright notice, and CC BY 4.0 requires keeping it along with the creator credit the licensor asks for. The same applies to the source dataset titles and Hugging Face identifiers.
• The README sentence "Enes Deniz / AltaySec licenses the new annotations … under CC BY-SA 4.0." This names the licensor of the contribution, which is a different thing from credit. @3nesdeniz, can you confirm whether you license these annotations personally? If so, a follow-up can change it to "Enes Deniz licenses …".

Verification

Run with #89's checks against this branch:

• Schema, DSGAI, identifier, anonymization and duplicate checks on promptinj_dataextraction_testcases: 0 errors, 0 warnings. The "no schema" warning for the aggregate file is gone.
• All 454 case IDs across the three collections are unique.
• tr_turkish_conversation_prompt_injection_pairs/validate.py passes. It cross-checks IDs and prompt similarity against the other collections.

Checklist

• Validation scripts pass
• Entries reference at least one DSGAI ID
• One entry per file

🤖 Generated with Claude Code