Add UK wealth decile impact outputs

[anth-volk]

Fixes #353
Fixes #354

Summary

• Adds decile_variable support to calculate_decile_impacts(...), allowing callers to measure changes in one variable while grouping by a precomputed decile variable.
• Adds UK wealth_decile_impacts and intra_wealth_decile_impacts to economic_impact_analysis(...).
• Documents the new helper argument, UK-only analysis fields, and the additive decile_variable dataframe column.

Testing

• uv run --extra dev python -m pytest tests/test_intra_decile_impact.py tests/test_uk_analysis.py -q
• uv run --extra dev ruff check src/policyengine/outputs/decile_impact.py src/policyengine/tax_benefit_models/uk/analysis.py tests/test_intra_decile_impact.py tests/test_uk_analysis.py
• uv run --extra dev ruff format --check src/policyengine/outputs/decile_impact.py src/policyengine/tax_benefit_models/uk/analysis.py tests/test_intra_decile_impact.py tests/test_uk_analysis.py

Notes

• This intentionally does not update policyengine-api-v2; API response mapping for wealth_decile and intra_wealth_decile will be handled separately.
• calculate_decile_impacts(...).dataframe now includes an additive decile_variable column so consumers can distinguish income-derived deciles from precomputed grouping variables.

[anth-volk]

@vahid-ahmadi I’d especially like your review on whether the UK wealth-decile formula/usage here is correct and typical: the new outputs measure household net income impacts grouped by household_wealth_decile, where household_wealth_decile comes from the UK package’s household wealth ranking.

[PavelMakarchuk]

Approving on methodology — verified that the relative_change formula (mean of per-household ratios) and the implicit negative-decile exclusion via mask == self.decile are both pre-existing patterns from #180 and the earlier IntraDecileImpact work, not introduced here. Wealth-decile outputs will be consistent with the income-decile outputs already produced by economic_impact_analysis.

Two non-blocking suggestions for follow-up:

1. Test brittleness in test_uk_analysis.py — the exact-equality kwargs assertions (e.g. assert decile_calls[0] == {"baseline_simulation": baseline, "reform_simulation": reform}) will break the next time anyone adds a parameter to calculate_decile_impacts. Prefer containment-style assertions (assert kwargs["decile_variable"] == "household_wealth_decile").

2. Schema asymmetry — UK's PolicyReformAnalysis now has wealth_decile_impacts and intra_wealth_decile_impacts as required fields while the US model does not, so generic consumers must branch on country. Worth a follow-up to decide whether to unify the schema with Optional[...] = None defaults.

Also worth noting separately (not for this PR): the DecileImpact.relative_change formula diverges from the legacy policyengine-api aggregate Σ(change)/Σ(baseline) measure. This is a pre-existing divergence from #180, but it will become user-visible whenever economic_impact_analysis gets wired into api-v2.