Publish US releases as a single national H5

[MaxGhenis]

Summary

• stop writing one-H5-per-state and one-H5-per-congressional-district artifacts from the US fiscal release builder
• keep the release artifact surface to a single national populace_us_2024.h5 plus diagnostics, manifests, reform validation, demographics, and source coverage
• reject US release manifests that try to publish states/ or districts/ artifact keys
• remove the now-obsolete regional artifact writer CLI and tests

Calibration context

This follows the sparse-default 57k full-target-surface work. The completed full-surface selected-H5 run used 32,637 fiscal targets and 57,234 nonzero households, with:

• initial loss: 0.045060887932777405
• final loss: 0.04423309330088607
• fraction within 10%: 0.7624168888071821
• release gates: no failures in calibration_diagnostics.json

A national-only release rerun is in progress from the same 57,240-household selected H5 and reconstructed warm-start weights. It intentionally does not write state or district files; API and policyengine.py consumers should use the single national file and region filters.

Verification

• git diff --check
• uv run ruff check tools/build_us_fiscal_refresh_release.py packages/populace-data/src/populace/data/contract.py packages/populace-build/tests/test_us_fiscal_refresh_builder.py packages/populace-data/tests/test_contract.py
• uv run ruff format --check tools/build_us_fiscal_refresh_release.py packages/populace-data/src/populace/data/contract.py packages/populace-build/tests/test_us_fiscal_refresh_builder.py packages/populace-data/tests/test_contract.py
• uv run pytest packages/populace-build/tests/test_us_fiscal_refresh_builder.py packages/populace-data/tests/test_contract.py

Consumer spot checks on current upstreams:

• policyengine-api origin/master: default US dataset handling already leaves default/national resolution to certified metadata/filters; focused tests passed.
• policyengine.py origin/main: US state and congressional district regions already use dataset_path=None plus row filters on the national Populace file.

[MaxGhenis]

National-only release follow-up from the local run currently validating:\n\n- Output contract: single root artifact only so far (populace_us_2024.h5 plus calibration NPZ/diagnostics); no states/ or districts/ artifacts are being written.\n- Calibration diagnostics from out/sparse-default-release-20260701/refit-selected-current-surface-b788afc-national-only-20260701/releases/populace-us-2024-sparse-l0-refit-57k-b788afc-national-only-20260701/calibration_diagnostics.json:\n - initial_loss: 0.04423392191529274\n - final_loss: 0.04393288697745081\n - fraction_within_10pct: 0.7672886601096914\n - n_nonzero: 57196\n - gate_failures: null\n- Full target surface:\n - n_targets: 32637\n - n_records: 57240\n - constraint matrix: 32637 x 57240, 6186732 nonzero entries\n - surface sha256: bc3e2dc57edf7795da6cab776554806bb426ffd449377e36af7348b141f5b2f1\n - names sha256: be5a1fd8879e12df6e3c90884cf5cf8f1170f9b06f29d74c32720ae3bbcfe50b\n - values sha256: 1725eed9db538c89afe628317aa6d5616f60459477db3a2f49c27dfcba12922e\n\nPublication is still pending the slow post-calibration reform-validation/manifests phase. I am not treating this as live until reform_validation.json, demographics.json, us_source_coverage.json, build_manifest.json, and release_manifest.json exist and the publish contract passes.