Make 2025 ACA take-up override state-aware

[daphnehanse11]

Summary

• make the 2025 ACA post-calibration take-up override use checked-in state APTC enrollment targets when available
• when state spending targets are available, select ACA takers by PTC per covered person so the final vector targets both enrollment and assigned PTC spending
• allow the override to add or remove tax-unit ACA take-up by state, instead of only topping up nationally
• add policyengine_us_data.calibration.diagnose_aca_state_targets for reusable staging-H5 ACA diagnostics
• bump policyengine-us to 1.680.0 so generated data includes the Basic Health Program ACA eligibility fix from policyengine-us#8127

Closes #764.

Test plan

• .venv/bin/python -m pytest tests/unit/test_enhanced_cps.py tests/unit/calibration/test_unified_calibration.py::TestAcaTakeupTargeting -q
• .venv/bin/python -m ruff check policyengine_us_data/utils/takeup.py policyengine_us_data/datasets/cps/enhanced_cps.py policyengine_us_data/calibration/diagnose_aca_state_targets.py tests/unit/test_enhanced_cps.py tests/unit/calibration/test_unified_calibration.py
• .venv/bin/python -m ruff format --check policyengine_us_data/utils/takeup.py policyengine_us_data/datasets/cps/enhanced_cps.py policyengine_us_data/calibration/diagnose_aca_state_targets.py tests/unit/test_enhanced_cps.py tests/unit/calibration/test_unified_calibration.py
• all-state staging diagnostic on cached 1.89.1_a2f3bb36_20260430_205113 state H5s under policyengine-us 1.680.0

Smoke diagnostic

Applied the updated selector to cached staging state H5s under policyengine-us 1.680.0:

State	Enrollment target	Selected	Enrollment error	Assigned PTC spending error
NY	118,526	118,542	+0.01%	+0.10%
MN	91,520	91,517	-0.004%	-2.68%
FL	4,472,539	2,961,537	-33.78%	-70.09%
TX	3,713,574	3,713,560	-0.0004%	-56.26%

All-state diagnostic output saved locally to /tmp/aca_diag_all_spending_best.csv:

• 50/51 states completed; NM still fails with an unrelated non-ASCII decode error in the staging H5.
• Only DC (-3.50%) and FL (-33.78%) are more than 1% off on adjusted enrollment among completed states.
• FL is a hard support ceiling: the file has 2.96M potential positive-PTC people against a 4.47M enrollment target.
• NY and MN, the alarming BHP states, are no longer over-allocated after combining policyengine-us#8127 with the state/spending-aware selector.
• Many states still have low assigned PTC spending even after high-subsidy selection, especially FL/TX/GA. That points to a separate target/model-dollar issue rather than take-up allocation; see Replace per-state CMS APTC target with IRS SOI Net Premium Tax Credit (A11560/N11560) #805 for the target-definition side.