Constituency and local-authority simulations get no UC deductions or private school attendance #1902

uc_deduction_random_draw, uc_deduction_type_random_draw and attends_private_school choose between microdata imputation and single-household defaults by testing whether total weight is below 1e6:

• policyengine_uk/variables/gov/dwp/universal_credit/deductions/uc_deduction_random_draw.py: benunit_weight.sum() < 1e6 gives a draw of 1.0, so no UC deductions.
• policyengine_uk/variables/gov/dwp/universal_credit/deductions/uc_deduction_type_random_draw.py: the same test.
• policyengine_uk/variables/contrib/labour/attends_private_school.py: household_weight.sum() < 1e6 gives no private school attendance.

policyengine.py builds constituency and local-authority simulations by filtering rows from the national dataset:

• src/policyengine/countries/uk/regions.py uses RowFilterStrategy on constituency_code_oa / la_code_oa.
• src/policyengine/utils/entity_utils.py filter_dataset_by_household_ids keeps rows without rescaling weights.
• src/policyengine/tax_benefit_models/uk/model.py run() wraps the filtered tables in UKSingleYearDataset and calls Microsimulation.

A constituency carries roughly 8k to 380k benefit units of weight, so every constituency and local-authority run gets household defaults. Those runs model no UC deductions at all and no private school attendance, including under a private school VAT reform. Northern Ireland as a filtered region carries 0.996m benefit units of weight (enhanced FRS 2024-25), just under the threshold.

#1899 adds Simulation.built_from_dataset, which is true for any simulation built from data. It stays true for a subset filtered from the data, so these variables can gate on how the simulation was built rather than on its weight.