Add UK detailed budget program statistics

[anth-volk]

Fixes #352

Summary

• Expand UK ProgramStatistics coverage for detailed-budget parity by adding fuel_duty, state_pension, ni_employer, and tax_credits.
• Add those UK variables to the default model output materialization list so analysis outputs can calculate them without caller-supplied extra_variables.
• Validate UK program-statistics configuration before running analysis, with clearer errors for missing model variables or missing materialized outputs.
• Add focused UK program-statistics tests, including a public economic_impact_analysis() wiring test with non-program outputs stubbed.

Notes

• tax_credits intentionally overlaps with working_tax_credit and child_tax_credit; downstream budget adapters should choose the rows they need rather than summing all UK program-statistics rows.
• This PR does not add a full UK integration simulation run. The tests use mocked output datasets to keep PR runtime bounded.

Tests

• uv run --extra dev python -m pytest tests/test_uk_program_statistics.py tests/test_us_program_statistics.py -q
• uv run ruff check tests/test_uk_program_statistics.py src/policyengine/tax_benefit_models/uk/analysis.py src/policyengine/tax_benefit_models/uk/model.py
• uv run ruff format --check tests/test_uk_program_statistics.py src/policyengine/tax_benefit_models/uk/analysis.py src/policyengine/tax_benefit_models/uk/model.py

[anth-volk]

@vahid-ahmadi I requested your review here. Could you specifically check whether the UK programs added in this PR align with the programs currently in use in the UK model/reporting flow, especially the tax_credits row and its relationship to working_tax_credit and child_tax_credit? The PR currently documents that these rows overlap and expects downstream adapters to choose the row set they need rather than summing all of them.

[PavelMakarchuk]

Approving. The four new programs (fuel_duty, state_pension, ni_employer, tax_credits) bring economic_impact_analysis(...).program_statistics to parity with policyengine-app's UKDetailedPrograms schema (src/schemas/societyWideModules.js). Entities, is_tax flags, and entity_variables wiring all line up. ProgramStatistics.run() itself is unchanged — this is pure additive wiring plus a pre-flight validation helper.

Test pattern is a nice step up from #358: the mocked simulation builds real MicroDataFrame / UKYearData / PolicyEngineUKDataset objects, with monkeypatching limited to Simulation.ensure and the unrelated non-program outputs.

Non-blocking suggestions for follow-up:

1. Hoist resolve_entity_variables(simulation) out of the program loop in _validate_program_statistics_config — it currently runs once per program × 2 simulations and returns the full dict each time.
2. Tighten the UK_PROGRAMS type hint — dict[str, dict[str, bool]] or a TypedDict would self-document.
3. tax_credits overlap with working_tax_credit + child_tax_credit — matches the legacy app behavior (the app only consumes tax_credits from UKDetailedPrograms), but worth either dropping the disaggregated rows or marking them with an overlapping=True flag in the data structure rather than relying on the PR description comment alone. Today a downstream consumer that naively sums all program_statistics rows will double-count.
4. _format_missing_program_variables is a one-liner used once — could be inlined.
5. No example update in examples/ — small miss.
6. Schema asymmetry — wealth_decile_impacts/program-stats list remains UK-only on PolicyReformAnalysis. Same follow-up as flagged on #358.