Generate a reusable evaluation and safety baseline

[sergeliatko]

Summary

• generate activation and handoff fixtures that must be specialized before maintenance mode
• generate primary workflow, incomplete-input, and verification scenarios
• generate inert adversarial scenarios for instruction injection, destruction, and public disclosure boundaries
• add a stable eight-case evaluation registry across four generic coverage segments
• validate fixture linkage, registry completeness, maintenance AGENTS.md sections, and the canonical runtime map
• add testing and conditional threat-model guidance that distinguishes deterministic, behavioral, and human evidence

Validation

• npm.cmd run validate
• npm.cmd run package -- v0.0.0
• node --check scripts/validate-evaluations.mjs
• git diff --check

Closes #15