Generate complete ACS WIC participation and reject missing release inputs

[anth-volk]

Fixes #1155

Summary

• Generate ACS-only takes_up_wic_if_eligible after demographic transfer and before combining ACS with donors. Reuse the existing seeded/category-specific WIC generator; preserve donor decisions exactly.
• Require complete current-name boolean participation before calculation/export and validate the actual H5 values before packaging. A reviewed-null entry cannot fill missing WIC from the country engine's True default.
• Record generator provenance and hash-bound checked row counts in existing build/release reports.
• Add unit tests, H5 round-trip tests, country-model consumption tests, and separate dataset-owner rebuild instructions.

Related changes and remaining work

• Parent replacement tracking: Publish complete ACS-local WIC participation and remove the temporary .py assumption #1154. Do not close that issue when this code merges.
• Temporary .py compatibility: Temporarily handle missing ACS WIC participation inputs policyengine.py#566
• Separately prepared .py removal: Remove temporary ACS WIC assumption after replacement certification policyengine.py#563 (blocked until the actual replacement is published and certified).

The existing published dataset is not modified. A separate owner must perform the full rebuild, rebuild dependent calculations/calibration, satisfy all existing release checks, and publish a new immutable non-default local-area release. National default/latest remains unchanged.

Validation

• Focused ACS-generation, orchestration, legacy-entry-point, release-tool, sparse-materialization, and source-independence tests passed. Existing tests requiring the optional pinned 164 MB aggregate feed were skipped because that local fixture is unavailable.
• 11 H5/country-engine workflow tests passed, including complete flag consumption and multi-batch reads.
• 8 focused packaging/default-fill/export regression tests passed.
• Repository-wide CI Ruff check, changed-file formatting, test-directory registration, and the existing graph-acceptance verification passed.
• Repository-wide optional formatting check reports existing unrelated files; these were not rewritten. Current CI runs Ruff lint, not that format check.
• No full simulation-executor test suite, real dataset rebuild/publication, paid compute, deployment, environment-variable provisioning, or database changes.

Fixed-format H5 row reads may still unpickle entire object blocks; the documentation does not promise a strict memory bound for that historical format.