Certify policyengine-us 2.15.17 in the US bundle

[MaxGhenis]

Problem

The certified US bundle in policyengine 6.1.2 pins policyengine-us==2.2.1 and policyengine-core==3.32.5. The newest policyengine-us on PyPI is 2.15.17, which requires policyengine-core>=3.32.8.

2.2.1 still has SNAP rounding defects that later releases fix. PolicyBench recomputed its 1,984 US household reference outputs (tax year 2026) under both versions:

• 22 outputs disagree: 13 SNAP, 8 federal_income_tax_before_refundable_credits and 1 state income tax.
• Under 2.2.1, 12 of 19 positive annual SNAP amounts are not whole dollars. One NJ household gets 15,309.53 under 2.2.1 and 15,273 under 2.15.17.
• Under 2.15.17, all 18 positive SNAP amounts are whole dollars.

PolicyBench records whether its reference model matches the policyengine.py certified bundle. Until policyengine.py certifies a current policyengine-us, every reference it builds on the newest model is flagged as uncertified.

Constraint

The certified data release populace-us-2024-spm-20260915 was built with policyengine-us 2.2.1. Its release manifest's publisher claim covers >=2.0.0,<2.3. Under the certification gate (docs/engineering/skills/data-certification.md), 2.15.17 needs either a new data build or a published compatibility claim for it.

Plan

1. Build the evidence that the certified H5 means the same thing to 2.15.17 as to 2.2.1:
   • stored-input audit, covering renamed, removed and now-computed inputs and enum members;
   • population aggregates under both versions, with every move explained;
   • the test suite on the pinned pair.
2. If the evidence holds, append a policyengine-us==2.15.17 claim to the release manifest, as Bump to policyengine-uk 2.90.2 and revert the UK certified default to enhanced FRS 2024-25 #491 and Bump policyengine-uk to 2.102.3 for the FYE 2025 absolute poverty line #536 did for UK.
3. Certify with scripts/bundle.py certify-data --country us --data-producer populace --model-version 2.15.17, and bump policyengine-core to 3.32.8.