Add TANF take-up to microdata construction

[MaxGhenis]

Summary

Adds state-level TANF takeup rates to the microdata construction pipeline, following the same pattern as Medicaid and SNAP. This is the data-side prerequisite for removing the tanf_reported short-circuit from policyengine-us (#7400).

Closes #535

Changes

New file: parameters/take_up/tanf.yaml

• State-level takeup rate priors for all 50 states + DC
• Derived from ACF FY2023 TANF caseload data (average monthly families by state) divided by Census ACS 2023 families in poverty
• Scaled ~2x to approximate takeup among eligible families (eligible pool ≈ 50% of poverty pool due to asset tests, work requirements, time limits)
• Rates range from 0.03 (AR) to 0.91 (CA), reflecting massive state-level variation in TANF accessibility

Modified: datasets/cps/cps.py

• Load TANF state-specific rates via load_take_up_rate("tanf", ...)
• Map SPM units to states (via household → state lookup), reused by Medicaid block
• Assign takes_up_tanf_if_eligible boolean to each SPM unit using deterministic seeded RNG

Calibration

The existing tanf: 9e9 hardcoded target in pull_hardcoded_targets.py will calibrate total TANF spending. The takeup priors determine which households receive TANF in the microsim; calibration adjusts weights to match the $9B national target.

Companion work needed (policyengine-us)

1. Add takes_up_tanf_if_eligible variable (bool, SPMUnit, default_value = True)
2. Wire all 39 state TANF programs into tanf.py (currently only 4 of 39)
3. Apply takeup in TANF formula: value * takes_up
4. Remove the tanf_reported short-circuit

See policyengine-us#7400.

Test plan

• tanf.yaml loads correctly via load_take_up_rate("tanf", 2023)
• add_takeup() assigns takes_up_tanf_if_eligible to all SPM units
• State mapping produces correct rates (spot-check CA ≈ 0.91, TX ≈ 0.04)
• Existing takeup assignments (SNAP, Medicaid, etc.) unaffected

🤖 Generated with Claude Code