Certify latest Populace US data release

[MaxGhenis]

Summary

Certifies the latest Populace-US 2024 release in policyengine.py:

• points the bundled US release manifest to populace-us-2024-f0af251-703bd81a565c-20260620T201958Z
• pins the certified dataset sha256 to 16be6338f9d0b3c339883dae59949e995663b64cf145de6728b3dd0f916c5d5f
• regenerates the US TRACE TRO sidecar for the new release
• fixes same-process TRO regeneration to clear cached manifest readers before rebuilding the sidecar
• updates tests that intentionally pin the certified US release id

Notes

The certified Populace build was produced with policyengine-us==1.729.0, which already matches the current .py US extra. The release manifest and supplemental diagnostics are reachable on Hugging Face and source coverage is included.

The published reform_validation.json is a diagnostic artifact. The recovery build skipped full out-of-sample reform microsimulation after the full validation path OOMed, so that diagnostic contains null Populace effects for out-of-sample rows. This PR does not treat that diagnostic as a hard certification gate; the dataset, manifest, source coverage, and runtime compatibility gates pass. We should follow up with a memory-bounded reform validation runner.

Validation

• uv run --all-extras python -m pytest -> 643 passed, 52 warnings
• uv run ruff format --check .
• uv run ruff check .
• git diff --check
• Runtime smoke: managed_microsimulation() resolves populace-us-2024-f0af251-703bd81a565c-20260620T201958Z, loads 75,112 households, and exposes the new data build id in policyengine_bundle
• Live HEAD checks returned 200 for calibration_diagnostics.json, reform_validation.json, and us_source_coverage.json
• /cycle read-only review found no actionable findings