Set the mixed-age couple Pension Credit saving at build time, from survey-year ages #519

What is needed

PolicyEngine/policyengine-uk#1940 adds the benefit-unit input has_mixed_age_couple_pension_credit_saving: whether a mixed-age couple keeps Pension Credit or pension-age Housing Benefit under the SI 2019/37 article 4 saving (entitled as that couple on 14 May 2019 and ever since).

The dataset does not supply it, so the model falls back on a default formula. That default requires, among other things, that the older member's birth_year is 1954 or earlier (the parameter gov.dwp.pension_credit.mixed_age_couples.saving_latest_birth_year).

Why the default drifts

• birth_year is computed as the model year minus age (policyengine_uk/variables/household/demographic/birth_year.py).
• Both reviews of #1940 (vahid-ahmadi, 1 October 2026; the Codex review of 2 October 2026) report that the Enhanced FRS holds each person's age fixed across model years. I have not re-checked that in the build.
• If so, the computed birth year moves forward one year per model year, and each year one more cohort of couples that really hold the saving falls outside the 1954 gate. Whether a couple was protected on 14 May 2019 is a historical fact and should not change with the model year.
• The PR body reports the count of Pension Credit eligible saved couples falling from 24.1k (2025) to 19.9k (2026), which is consistent with this.

Proposal

Set has_mixed_age_couple_pension_credit_saving once at build time, from survey-year facts:

• the couple is mixed-age in the survey year, by survey-year ages;
• the older member was born on or before 5 February 1954 (survey year minus survey age, or the exact date of birth if the FRS field allows);
• the couple reports Pension Credit, or the pension-age member reports Housing Benefit without Income Support, income-based JSA or income-related ESA;
• no reported Universal Credit.

These are the same conditions the model's default uses, evaluated once in the survey year, so the flag stays fixed in later model years.

Notes

• Merging any uk-data PR publishes a dataset version, so this should go into a planned release.
• Until this lands, long-horizon estimates of the saving understate it; #1940's body says so.

Raised from the reviews of PolicyEngine/policyengine-uk#1940.