Weight US fiscal calibration loss by target value

[MaxGhenis]

Summary

• add optional target row weights to the calibration relative-error objective
• keep unweighted calibration as the default API behavior
• make the US fiscal refresh builder weight target rows by normalized absolute target value, so national dollar totals are not one vote among thousands of rows
• record the fiscal refresh weighting strategy in calibration diagnostics build metadata

Why

The broken release exposed that an unweighted row average is not a sufficient solve priority for a target surface dominated by many state-level rows. Hard gates now prevent critical misses from certifying; this PR changes the optimizer objective so large national fiscal targets receive proportionate pull during the solve.

Validation

• PYTHONPATH=packages/populace-calibrate/src:packages/populace-frame/src uv run --python 3.13 --with pytest --with numpy --with pandas --with torch --with scipy python -m pytest packages/populace-calibrate/tests/test_solve.py -q
• PYTHONPATH=packages/populace-build/src:packages/populace-calibrate/src:packages/populace-frame/src:packages/populace-fit/src:packages/populace-data/src uv run --python 3.13 --with pytest --with numpy --with pandas --with packaging --with h5py --with torch --with scipy python -m pytest packages/populace-build/tests/test_gates.py packages/populace-build/tests/test_us_fiscal_refresh_builder.py -q
• uv run --python 3.13 ruff check packages/populace-calibrate/src/populace/calibrate/solve.py packages/populace-calibrate/tests/test_solve.py packages/populace-build/tests/test_gates.py tools/build_us_fiscal_refresh_release.py packages/populace-build/tests/test_us_fiscal_refresh_builder.py
• uv run --python 3.13 ruff format --check packages/populace-calibrate/src/populace/calibrate/solve.py packages/populace-calibrate/tests/test_solve.py packages/populace-build/tests/test_gates.py tools/build_us_fiscal_refresh_release.py packages/populace-build/tests/test_us_fiscal_refresh_builder.py && git diff --check