Drop formulaic SPM outputs from exported H5s

[MaxGhenis]

Summary

• centralize the formulaic SPM poverty/geographic-adjustment variables that should not be persisted as dataset leaf inputs
• apply that filter to cloned household and local-area H5 export paths
• extend tests so person-level poverty formula outputs are excluded alongside thresholds and geographic adjustments

Tests

• env -u UV_FROZEN uv run ruff check policyengine_us_data/calibration/formulaic_inputs.py policyengine_us_data/calibration/entity_clone.py policyengine_us_data/calibration/publish_local_area.py tests/unit/calibration/test_formulaic_inputs.py tests/unit/calibration/test_entity_clone.py tests/unit/test_extended_cps.py
• env -u UV_FROZEN uv run pytest tests/unit/calibration/test_formulaic_inputs.py tests/unit/calibration/test_entity_clone.py tests/unit/test_extended_cps.py -q