Drop raw-source and build-intermediate columns from the published populace-us artifact

[MaxGhenis]

Summary

The published populace-us artifact should store PolicyEngine-recognized inputs plus explicit structural/geo columns, not raw source-survey fields or build scratch columns.

Drop (raw source + build intermediates)

• Raw CPS ASEC columns: A_AGE, A_FTPT, WSAL_VAL, SEMP_VAL, WC_VAL, DIS_VAL1, etc.
• Raw SCF columns such as scf_net_worth, scf_bank_account_assets, and other scf_* source fields once the PE inputs they feed are set.
• Build scratch: _half, _orig_household_id, *_is_puf_clone, new_tax_unit_id, and redundant intermediate weights if they duplicate calibrated weights.

Keep

• Geo identifiers used by the local-area pipeline: block_geoid, tract_geoid, congressional_district_geoid, cbsa, etc.
• Structural keys and calibrated weights.

Acceptance criteria

• The export surface has an explicit structural/geo allow-list.
• Raw source and scratch columns are removed upstream before final artifact writing.
• A gate fails when an artifact contains columns outside PE variables plus the allow-list.
• Tests cover a raw-source column, a scratch column, and a retained geo/structural column.

[MaxGhenis]

PE-US-data backlog migration note from #46: this export-surface issue should retain the lessons from PolicyEngine/policyengine-us-data#444, #829, #348, #349, #814, #441, and #149 where they still identify formula-owned exports, entity-ID propagation, filing-status, or naming problems. Do not import old eCPS parity behavior; Populace's final artifact should be PE inputs plus explicit structural/geo allow-list.