Rollback US Populace default and gate critical targets

[MaxGhenis]

Fixes #421.

Summary

• Restores the US bundled Populace default from the failed 2026-06-17 release to the prior 2026-06-16 release.
• Adds a Populace certification gate in the release-bundle refresh path for critical fiscal targets:
  • federal income tax liability amount: max 5% absolute relative error
  • federal income tax liability returns: max 10% absolute relative error
  • Social Security benefit payments: max 5% absolute relative error
• Requires Populace releases to include calibration_diagnostics before .py certification can refresh them.

Verification

• env -u UV_FROZEN uv run ruff check src/policyengine/provenance/bundle.py tests/test_bundle_refresh.py tests/test_release_manifests.py tests/test_models.py
• POLICYENGINE_SKIP_COUNTRY_IMPORTS=1 env -u UV_FROZEN uv run --with pytest python -m pytest --noconftest tests/test_bundle_refresh.py -q (27 passed)
• env -u UV_FROZEN uv run --extra us --extra uk --with pytest python -m pytest --noconftest tests/test_certify_data_release.py tests/test_bump_version.py tests/test_release_manifests.py tests/test_trace_tro.py tests/test_models.py -q (134 passed)
• Live HF check: bad populace-us-2024-f0d2ef6-09292aa0d5db-20260617T032307Z fails the new gate on all three critical targets.
• Live HF check: restored populace-us-2024-a912aea-76666318a202-20260616T175345Z passes the new gate.