Treat Medicaid spending as validation-only target

[MaxGhenis]

Summary

• classify CMS NHE Medicaid spending as validation-only instead of an active hard calibration target
• keep Medicaid and CHIP enrollment controls active
• move the CMS NHE package alias from active release sources to reviewed exclusions for this fiscal refresh

Root Cause

PolicyEngine-US computes medicaid spending from state spending parameters through person_weight-dependent Medicaid SLCSP denominators. Reweighting changes each enrolled person's assigned Medicaid cost, so the row is not a linear calibration target. The post-export audit caught this: calibration reported about $519B under frozen pre-calibration costs, while the exported H5 recomputed Medicaid at about $880B.

Validation

• env -u UV_FROZEN uv run --package populace-build --extra us python -m pytest packages/populace-build/tests/test_us_fiscal_targets.py packages/populace-build/tests/test_us_source_coverage.py packages/populace-build/tests/test_us_fiscal_refresh_builder.py -q
• env -u UV_FROZEN uv run --all-packages ruff check .
• env -u UV_FROZEN uv run --all-packages python -m pytest -q
• audited release build with tools/build_us_fiscal_refresh_release.py --audit-export-targets completed: populace-us-2024-f0d2ef6-45b320b7cee5-20260617T030430Z
• validate_release_dir(...) passed for that release
• local PolicyEngine-US smoke loaded the generated H5 and calculated income_tax, social_security, ssi, medicaid, medicaid_enrolled, and aca_ptc

Notes

The new audited build has 5,097 active hard targets: one fewer than before, exactly the invalid Medicaid spending row. CMS Medicaid enrollment and Medicaid+CHIP enrollment remain active (104 CMS Medicaid target rows). The CMS NHE source package is recorded in reviewed exclusions rather than claimed as active hard-target coverage.