CD vintage crosswalk is an unpinned build input; July-1 release surface not reproducible

Problem

The July-1 sparse release (populace-us-2024-sparse-l0-refit-57k-71a0887-national-only-20260701) compiled its 32,637-target surface using --congressional-district-vintage-crosswalk cd117_to_cd119_population_crosswalk.csv. The build manifest records the file's sha256 (a615539…) but the artifact itself was never published — it lived in a since-deleted local worktree. The release's target surface is therefore not reproducible: rebuilding with the same facts and flags cannot recover the exact 117th→119th population weighting.

Sibling of #271 (Ledger facts were unversioned in the manifest; fixed by the pinned consumer artifacts) — the CD vintage crosswalk is the remaining unpinned build input.

Suggested fix

• Publish build-input artifacts like the CD vintage crosswalk alongside the release (or in a pinned artifact repo) and record repo/revision, not just sha256.
• Regenerate cd117_to_cd119_population_crosswalk.csv from Census block population data (117th→119th, 5001700US → 5001900US), validated against the per-state 119th district counts recorded in us/fiscal_target_references.json hierarchy reconciliations (CA 52, TX 38, FL 28, NY 26, …), and pin it for the corrected-release rebuild.

Found while rebuilding the corrected sparse release: the current build compiles only the 5.5k national+state surface without the flag+crosswalk pair.

Start by inspecting the release build inputs and us/fiscal_target_references.json, then trace how cd117_to_cd119_population_crosswalk.csv is consumed for the congressional-district surface. Done means the regenerated crosswalk is validated against the recorded 119th district counts, published or pinned with repository and revision metadata, and used to reproduce the corrected sparse release.