Ground EMAP’s claims in checked-in artifacts

[noah-ing]

Summary

• replace causal and full-benchmark claims with a reproducible account of the 12 checked-in artifacts and their limits
• replace the paper and PDF with an artifact-grounded research note
• remove the misleading dashboard and server
• make generated-code evaluation fail closed when subprocess isolation fails
• add resource limits, regression coverage, pinned read-only CI, a license, and private-reporting guidance

Verification

• Reconciled all 12 JSON artifacts and README/paper totals
• Compiled every changed Python source
• Inspected the rendered two-page PDF
• Full GitHub CI pytest suite