Test: D4D for CM4AI (file mode, pre-staged inputs)

[realmarcin]

@d4dassistant Please create a D4D for CM4AI using file mode.

Dataset name: CM4AI
Input mode: file
Input location: data/sheets_d4dassistant/inputs/CM4AI/

The following 6 source files are pre-staged in the repo:

• 2024.05.21.589311v1.full.txt (CM4AI preprint)
• RePORT ⟩ RePORTER - CM4AI.txt (NIH RePORTER record)
• creativecommons_org_licenses-by-nc-sa_row15.txt (license terms)
• dataverse_10.18130_V3_B35XWX_row16.txt (Dataverse record 1)
• dataverse_10.18130_V3_F3TD5R_row19.txt (Dataverse record 2)
• doi_row3.json (DOI metadata)

Please run the file-mode workflow described in .github/workflows/d4d_assistant_create.md:

1. ./src/github/validate_prerequisites.sh --dataset CM4AI --mode file
2. Synthesize across all six sources (the value of file mode is multi-document synthesis).
3. Validate the generated YAML against the schema and run validate_d4d_completeness.py.
4. Open a PR with the generated CM4AI_d4d.yaml, metadata YAML, and HTML preview.

Test purpose: This is a deliberate end-to-end test of the GitHub D4D assistant in file mode. We plan to diff the output against the curated baseline at data/d4d_concatenated/curated/CM4AI_curated.yaml and against the existing claudecode_agent output as quality references.