Map disability benefit categories in UK data

[MaxGhenis]

Summary

• Add dataset-side mapping from reported PIP, DLA, and Attendance Allowance amounts to PE-UK category inputs using current-law PE-UK rates.
• Keep disability reported amounts as internal FRS/imputation intermediates, then strip them before saved public datasets.
• Update public transfer data to emit PIP category inputs directly and remove obsolete disability reported amount uprating rows.

Companion to PolicyEngine/policyengine-uk#1656.

Tests

• uv run ruff check policyengine_uk_data/datasets/disability_benefits.py policyengine_uk_data/datasets/frs.py policyengine_uk_data/datasets/create_datasets.py policyengine_uk_data/datasets/enhanced_cps.py policyengine_uk_data/datasets/imputations/frs_only.py policyengine_uk_data/datasets/imputations/income.py policyengine_uk_data/tests/test_disability_benefits.py policyengine_uk_data/tests/test_policybench_transfer.py
• uv run pytest -q policyengine_uk_data/tests/test_disability_benefits.py policyengine_uk_data/tests/test_frs_only_imputation.py policyengine_uk_data/tests/test_policybench_transfer.py::test_policybench_transfer_writes_only_valid_leaf_inputs policyengine_uk_data/tests/test_policybench_transfer.py::test_policybench_transfer_runs_uk_microsimulation
• PYTHONPATH=/private/tmp/pe-uk-pr1656 uv run pytest -q policyengine_uk_data/tests/test_disability_benefits.py policyengine_uk_data/tests/test_policybench_transfer.py::test_policybench_transfer_writes_only_valid_leaf_inputs policyengine_uk_data/tests/test_policybench_transfer.py::test_policybench_transfer_runs_uk_microsimulation
• uv run pytest -q policyengine_uk_data/tests/test_policybench_transfer.py

[MaxGhenis]

Addressed cross-PR review findings:

• Regenerated enhanced_cps_2025.h5 so the checked-in artifact now has pip_dl_category / pip_m_category and no PIP reported amount columns.
• Added a regression test that loads the checked-in H5 and verifies those columns.
• Recomputed is_disabled_for_benefits, is_enhanced_disabled_for_benefits, and is_severely_disabled_for_benefits after stage-2 disability amount imputation, with tests.
• Kept current-law PE-UK rates for category assignment and documented that choice in the mapper.

[MaxGhenis]

Small follow-up after checking DLA aggregate preservation: category assignment now uses the same baseline DWP thresholds that the removed PE-UK category formulas used. The disability flag recomputation still uses the current-law DWP rates, matching the existing UK-data FRS derivation. This should keep DLA/PIP/AA category assignment behavior aligned with the pre-PR model while still publishing concrete category inputs instead of reported amount intermediates.

[MaxGhenis]

Verified aggregate preservation on the cached private enhanced FRS. I converted the old reported-amount dataset into category inputs with this PR's mapper, then compared current PE-UK on the old dataset against PE-UK #1656 on the converted dataset. Totals are exactly identical for the affected disability benefits: DLA £4.449897bn, DLA self-care £2.936218bn, DLA mobility £1.513680bn, PIP £25.045953bn, AA £7.966628bn.