Fix negative income values

[vahid-ahmadi]

Summary

• Add np.maximum(0, ...) clipping to 5 income variables that previously allowed negative values
• Add test to validate all income variables are non-negative

Variables fixed

Variable	Line
employment_income	342
self_employment_income	368
tax_free_savings_income	372-380
savings_interest_income	393-395
dividend_income	396-411

Variables already protected (no changes needed)

• private_pension_income - filters (pension.penpay > 0)
• property_income - uses np.maximum(0, ...)
• maintenance_income - uses sum_positive_variables()
• miscellaneous_income - uses sum_from_positive_fields()

Test plan

• Run pytest policyengine_uk_data/tests/test_non_negative_incomes.py
• Verify all 9 income variables pass the non-negative check

[vahid-ahmadi]

This PR should fix the error @anth-volk raised.

[vahid-ahmadi]

@PolicyEngine Review this pr pls

[policyengine]

Code Review ✅

I've reviewed this PR and the implementation looks solid. Here's my assessment:

Changes Verified

Variables correctly fixed with np.maximum(0, ...):

• employment_income (line 342-343)
• self_employment_income (line 370-371)
• tax_free_savings_income (line 376-383)
• savings_interest_income (line 397-400)
• dividend_income (line 401-415)

Variables already protected (correctly identified in PR):

• private_pension_income: uses pension.penpay > 0 filter (line 347)
• property_income: uses np.maximum(0, ...) (line 430)
• maintenance_income: uses sum_positive_variables() helper (line 446)
• miscellaneous_income: uses sum_from_positive_fields() helper (line 464)

Test Quality

The test in test_non_negative_incomes.py is well-designed:

• Parametrised across all 9 income variables
• Provides clear error messages with actual minimum values
• Comprehensive coverage

Code Quality

✅ Consistent use of np.maximum(0, ...) pattern
✅ Maintains existing code style and structure
✅ No over-engineering or unnecessary changes
✅ Proper changelog entry added

Recommendation

Approve and merge. This PR correctly addresses the negative income issue whilst respecting existing patterns and helper functions. The implementation is clean and the test coverage is appropriate.