Add TANF administrative calibration targets

[MaxGhenis]

Summary

This PR adds TANF administrative calibration targets from HHS ACF FY2024 data and wires them into the US data build.

It does three things:

• adds a new TANF ETL that reads the official FY2024 ACF caseload and TANF/MOE financial workbooks
• loads national and state TANF recipient-family counts plus TANF cash-assistance spending into the calibration database and target config
• exposes baseline CPS liquid-asset inputs (bank_account_assets, stock_assets, bond_assets) so follow-up state-specific resource modeling can use explicit asset types instead of survey-reported placeholders

Details

The new ETL pulls:

• average monthly TANF recipient families from the FY2024 ACF caseload workbook
• cash-assistance spending from the FY2024 TANF/MOE financial workbook, using the narrower row that excludes relative foster care maintenance payments and adoption/guardianship subsidies

Those targets are added at:

• national tanf
• national spm_unit_count with domain_variable: tanf
• state tanf
• state spm_unit_count with domain_variable: tanf

The old hardcoded national TANF total is removed from etl_national_targets.py, and the remaining legacy fallback is aligned to the same FY2024 ACF cash-assistance total.

Scope boundary

This PR does not publish a generic spm_unit_assets aggregate. That would be too semantically broad for the current data, because this branch only exposes liquid-asset inputs and many downstream rules still interpret spm_unit_assets as broader countable resources.

It also intentionally does not add state-specific TANF take-up priors. Full-take-up profiling still leaves structural eligibility misses in some states, so inferred state take-up rates are not yet credible. That follow-up belongs after the remaining eligibility/application-month modeling work.

For now, this PR stays on the calibration/data side:

• administrative TANF count and spending targets
• person-level liquid-asset inputs for future state-specific counted-asset formulas

Related context:

• Add TANF take-up to microdata construction #535
• Fix TANF application-month income and assistance-unit modeling in CPS baseline policyengine-us#7988
• Fix TANF asset modeling and immigration timing policyengine-us#7989

Verification

• uv run pytest -q tests/unit/test_etl_tanf.py tests/unit/calibration/test_target_config.py tests/unit/calibration/test_loss_targets.py tests/integration/test_cps_generation.py tests/integration/test_database_build.py
• uv run ruff check policyengine_us_data/db/etl_tanf.py policyengine_us_data/storage/calibration_targets/pull_hardcoded_targets.py policyengine_us_data/utils/loss.py tests/unit/test_etl_tanf.py tests/unit/calibration/test_loss_targets.py tests/integration/test_cps_generation.py tests/integration/test_database_build.py

Two fresh independent review passes after the follow-up patch found no remaining actionable issues.

[vercel]

@MaxGhenis is attempting to deploy a commit to the PolicyEngine Team on Vercel.

A member of the Team first needs to authorize it.

[MaxGhenis]

Closing this fork-based PR in favor of #750 so the repository's full CI can run. The branch is now mirrored to \ with the same commits plus the upstream-branch workflow note in README.