Improve SSI disability imputation predictors

[MaxGhenis]

Summary

• use the six comparable CPS/SIPP disability difficulty questions as separate SSI disability criteria model predictors instead of broad SIPP-only disability aggregates
• export and impute only the PE-US-facing meets_ssi_disability_criteria variable, with the old SSA-screen alias removed
• keep difficulty-question fields as us-data construction-time predictors, not PolicyEngine-US variables, and drop them before final export
• align SIPP SSDI to CPS semantics by using annualized Social Security dollars when disability is the reported reason
• refresh person-level CPS-only attributes onto PUF clones through donor rematching unless they are structural, PUF-imputed, or already handled by the CPS-only QRF stage
• bump policyengine-us from 1.705.6 to current PyPI 1.705.11 for CI freshness

Dependency ordering

• Related PE-US PR: Add SSI disability criteria input policyengine-us#8398 keeps meets_ssi_disability_criteria as the runtime/net-income-tree concept while removing the old SSA-screen alias.
• This us-data PR now writes only meets_ssi_disability_criteria, so it does not depend on the removed alias.

Tests

• .venv/bin/python -m pytest tests/unit/datasets/test_sipp_ssi_disability.py tests/unit/calibration/test_source_impute.py tests/unit/test_extended_cps.py tests/integration/test_cps_generation.py -q
• .venv/bin/ruff format --check .
• .venv/bin/ruff check policyengine_us_data/datasets/sipp/sipp.py policyengine_us_data/datasets/sipp/__init__.py policyengine_us_data/datasets/cps/cps.py policyengine_us_data/calibration/source_impute.py policyengine_us_data/datasets/cps/extended_cps.py policyengine_us_data/utils/dataset_validation.py tests/integration/test_cps_generation.py tests/unit/datasets/test_sipp_ssi_disability.py tests/unit/calibration/test_source_impute.py tests/unit/test_extended_cps.py
• uv run python .github/scripts/check_policyengine_us_dependency.py --mode fail