Core policies 1c-1: one statement of the pet rules

Eval evidence for #17295. Both new rows fail on the code before this change and pass with it.

Row	Before (1b code), ×3	This PR, ×3
Field-level: the company says no pets, and the building's older record holds only a 50 lb limit. "Can I bring my 30 lb dog?" must get no pets.	0 / 3	3 / 3
One statement: the older "up to three pets" entry was folded into an answer of up to 2. "How many pets can I have?" must get 2, never 3.	0 / 3	3 / 3
Full Core Policies pack on the PR head (`71ecde5`)		27 / 27

What "before" got wrong

Field-level row. Before this change, Clara never said no pets. The building's older weight record outranked the company's whole answer, including whether pets are allowed at all.

One-statement row. The pack's context check stopped the run before Clara answered. It found two "Pets" statements in her context, the old entry and the new answer.

How it was run

The pack is evals/promptfooconfig-core-policies.yaml, run through scripts/eval-core-policies-subscription.ts on claude-sonnet-4-6.

Every input comes from the production readers, run over synthetic companies.

"Before" was the 1b code with only this PR's eval rows and fixtures copied in.