Gate US fiscal refresh base population scale

[daphnehanse11]

Summary

• Add a pre-calibration base-population gate for the US fiscal refresh builder.
• Fail early when the base H5's resolved person-weight mass is far from the Census benchmark while the release calibration uses mass='conserve'.
• Record the passing base-population gate in release manifests and calibration build metadata.

Validation

• uv run ruff check tools/build_us_fiscal_refresh_release.py packages/populace-build/tests/test_us_fiscal_refresh_builder.py
• uv run pytest packages/populace-build/tests/test_us_source_coverage.py packages/populace-build/tests/test_us_fiscal_refresh_builder.py -q
• Local cached current H5 gate check fails as expected: 56,918,517 weighted people vs 334,200,000 benchmark (-83.0%).

Refs #67
Refs #94

[daphnehanse11]

Updated this PR for the review finding:

• rebased onto current main
• resolved the builder conflict by preserving Redesign calibration loss as capped weighted MAPE #96/Weight fiscal calibration loss by target value basis #97's _fiscal_target_loss_weights and US_FISCAL_TARGET_LOSS_CAP
• kept the base-population gate and diagnostics/manifest plumbing intact

Verification:

• uv run pytest packages/populace-build/tests/test_us_fiscal_refresh_builder.py
• uv run ruff check tools/build_us_fiscal_refresh_release.py packages/populace-build/tests/test_us_fiscal_refresh_builder.py
• GitHub test/wheel checks are green for Python 3.13 and 3.14