Household-impact tests return 0 for expected-positive tax/benefit values on 2.88.0

Summary

In policyengine.py's [uk] / [dev] extras bumped to policyengine-uk==2.88.0, two household-impact integration tests assert positive tax / child-benefit values that now come back as 0.0:

• TestUKHouseholdImpact.test_single_adult_with_employment_income — assert 0.0 > 0 on expected positive tax
• TestUKHouseholdImpact.test_family_with_children — assert 0.0 > 0 on expected positive child benefit

Under the old policyengine-uk==2.74.0 pin, both passed.

Reproducer: https://github.com/PolicyEngine/policyengine.py/pull/280 — Test (3.13) / Test (3.14) jobs.

Hypothesis

Either:

• uk 2.74 → 2.88 legitimately changed the 2024 household calculation semantics (e.g. different policy year defaults, tax threshold shifts) — then the pypkg test expectations need updating, or
• There's a regression in uk 2.88.0 for these specific household compositions.

I don't know the uk tax-benefit details well enough to diagnose.

Not blocking

Doesn't block the 3.9-support work that's already merged — uk main CI passes on 2.88.0 because uk's own YAML tests don't exercise this specific pypkg household path.

Start with the [uk] and [dev] extras in policyengine.py and the two named TestUKHouseholdImpact tests. Reproduce the failures in the Test (3.13) and Test (3.14) jobs, then compare behavior between policyengine-uk 2.74.0 and 2.88.0 for both household compositions. Done means identifying whether the dependency change is intentional or a regression and restoring correct passing tests.