Add fine AGI bracket targets and calibration improvements

[baogorek]

Summary

• Add fine AGI bracket targets from IRS SOI data to improve high-income weight distribution:
  • State-level: stubs 9/10 from in55cmcsv ($500K–$1M, $1M+) — person count and AGI
  • National: Table 1.4 fine brackets ($500K–$10M+, 6 bins) — tax unit count and AGI
• Re-enable income_tax_positive in target_config — previously excluded at 22% error, but without it US.h5 overshoots to 54%. The AGI brackets should further reduce the error.
• Add net_worth ($160T, Federal Reserve SCF) — already in DB ETL, just not wired into config
• Add district-level SNAP household counts (ACS S2201) — complements existing state SNAP dollars and national CBO target
• Fix test_immigration_status_diversity — use weighted counts via MicroSeries groupby instead of unweighted np.unique record counts

Motivation

Report card comparison shows enhanced_cps_2024.h5 outperforming US.h5 on income tax (10.6% vs 54.3%), tips (0.1% vs 37.5%), real estate taxes (1.6% vs 52.6%), and rent (1.2% vs 50.2%). The fine AGI brackets constrain the high-income weight distribution, which should reduce cascading errors in income tax and self-employment income.

Test plan

• Full pipeline rebuild with updated DB and target_config
• Compare report card before/after for income_tax, self_employment_income, and poverty_rate
• Verify fine AGI bracket strata are created correctly in policy_data.db
• Integration test passes with weighted immigration status check

🤖 Generated with Claude Code

[MaxGhenis]

Looks good after the state AGI overlap fix.