docs: add deterministic multi-family admissibility benchmark documentation

[ProfRandom92]

Motivation

• Provide a focused, contributor-facing guide that explains the deterministic multi-family admissibility benchmark purpose, pipeline, invariants, regeneration, validation, and regression protections.
• Summary: Document deterministic multi-family admissibility benchmark behavior, pipeline, determinism guarantees, regeneration commands, validation commands, and regression protections.

Description

• Added a new benchmark doc at docs/benchmarks/multi_family_admissibility_benchmark.md that includes a Mermaid pipeline diagram, the current fixture families, the four standard degradation levels, determinism guarantees, regeneration commands, validation commands, regression protections, and non-goals; no implementation, fixture, artifact, CI, or package changes were made.
• Changed files: docs/benchmarks/multi_family_admissibility_benchmark.md.

Testing

• Ran npm run check, which executes layout, typecheck, validation, build, and the full Python test suite, and the command completed successfully.
• The test run included pytest where all tests passed (213 passed), and the repository checks are green.
• Risks: Low; this is a docs-only change with no code or artifact modifications, and Next: optionally cross-link this benchmark doc from broader benchmark index docs in a follow-up docs-only PR.

---

Codex Task

[gemini-code-assist]

Code Review

This pull request introduces documentation for the Deterministic Multi-Family Admissibility Benchmark, outlining its purpose, pipeline, and regression protections. The review identified an inconsistency in the regression protection logic where the stated requirement for distinct behavior between 'mild' and 'moderate' levels conflicted with the use of a non-strict inequality operator in the documentation. A correction was suggested to enforce strict inequality across all degradation levels.

[gemini-code-assist]

The progression formula should use a strict inequality (>) between mild and moderate to be consistent with the requirement stated on line 86 that these behaviors must be distinct. Using >= allows for identical scores, which contradicts the stated goal of ensuring distinct behavior across these levels.

Suggested change

	- `baseline > mild >= moderate > severe`
	- baseline > mild > moderate > severe