Add TANF take-up to microdata construction

[MaxGhenis]

Summary

TANF currently uses a tanf_reported short-circuit in policyengine-us (#7400) rather than the modeled-with-takeup pattern used by SNAP, EITC, Medicaid, and SSI. Now that 39 states have TANF implementations, we should add TANF takeup to the data package.

Proposed change

1. Add parameters/take_up/tanf.yaml with state-level takeup rate priors derived from ACF FY2023 caseload data and Census poverty estimates
2. Add TANF takeup assignment in cps.py:add_takeup() following the Medicaid state-specific pattern
3. Companion PR in policyengine-us to add takes_up_tanf_if_eligible variable and wire it into TANF formulas

Calibration targets

Already exists: tanf: 9e9 in pull_hardcoded_targets.py (line 19) — $9B total TANF cash assistance from HHS/ACF

Proposed additions (for future calibration improvement):

• State-level TANF caseload targets from ACF FY2023 data (826K average monthly families nationally)
• State-level total cash assistance spending from ACF FY2023 financial data ($8.3B nationally on basic assistance)

Takeup rate priors

Using TANF-to-poverty ratios from ACF caseload data / Census ACS poverty estimates, scaled up ~2x to approximate takeup among eligible (since eligible families ≈ 50% of families in poverty due to asset tests, work requirements, time limits). These are priors that will shift during calibration.

National variation is extreme: AR ~3%, TX ~4% to CA ~90%, OR ~60%.

Sources (all primary/gov)

• ACF FY2023 TANF Caseload Data
• ACF FY2023 TANF Financial Data
• Census Bureau 2023 ACS 1-Year Estimates (Table S1702, families in poverty by state)

Related

• Move TANF reported value logic from policy model to data package policyengine-us#7400: Move TANF reported value logic from policy model to data package
• Remove is_in_microsim checks from benefit variables policyengine-us#7331: Remove is_in_microsim checks from benefit variables