Fix leakage in constrained non-household amount targets

[juaristi22]

Fixes #763

Summary

The unified calibration matrix builder was leaking non-qualifying dollars into constrained amount targets whose native entity is not household. Before this change, the builder would pre-aggregate those targets to household, evaluate constraints at person level, promote the result to a household-level any(), and then assign the full household amount whenever any qualifying person was present. In mixed-eligibility households, that overstated rows such as filer-only total_self_employment_income.

What This Changes

This PR changes constrained non-household amount targets to be filtered at their native entity before they are rolled up to household contributions in the matrix.

Concretely, the builder now:

• records each target variable's native entity (person, tax_unit, spm_unit, or household)
• precomputes non-household amount targets both at household level and at their native entity level during state and county precomputation
• builds reusable entity-to-household and person-to-entity index maps for person, tax-unit, and SPM-unit targets
• assembles clone-level entity arrays alongside household and person arrays
• updates _calculate_target_values_standalone() so baseline constrained amount targets with non-household native entities select qualifying entity amounts first and only then sum those qualifying amounts back to household

This preserves the existing behavior for household-native amount targets and leaves count-target semantics unchanged.

Effect On Issue #763

For the motivating case from the issue:

• Person A: total_self_employment_income = 10,000, filer
• Person B: total_self_employment_income = 5,000, non-filer

The constrained filer-only target now contributes 10,000 instead of the previous 15,000. More generally, any constrained amount target whose native entity is person, tax_unit, or spm_unit now includes only qualifying entity amounts.

Regression Coverage

This PR adds regression tests covering:

• person-entity amount targets filtering before household rollup
• tax-unit amount targets contributing once per qualifying unit
• household-native amount targets preserving current household any() semantics
• person-entity amount targets remaining scoped to the correct household
• SPM-unit amount targets contributing once per qualifying unit
• count targets preserving existing entity-aware behavior

[anth-volk]

Integ tests take long enough that I ported two minor CI/CD fixes from #749 to this PR.

[anth-volk]

Taking liberty of merging. Error is spurious ACA issue previously identified.