Add county-crosswalked FNS SNAP data as a second congressional-district benchmark

Problem

CD-level SNAP validation currently has one lens: ACS S2201 district household estimates (#240, PR #241) — a survey estimate with its own sampling error. Calibrating or validating against a single noisy benchmark launders that noise into confident-looking district results.

Proposal

USDA FNS publishes county-level SNAP participation and issuance (bi-annual). Population-weighted county→CD crosswalk (Census blocks or MCDC geocorr) yields an administrative district benchmark independent of ACS sampling error.

1. Ledger-side: ingest the FNS county bi-annual file as a source package
2. Crosswalk counties to current-vintage CDs (the #277 block ladder / #205 vintage translation machinery is directly reusable)
3. Add as a second validation_only family beside snap_local_proxy
4. Promotion rule: CD SNAP targets stay validation-only until the two independent benchmarks corroborate each other district-by-district; divergence localizes either thin raw support or a real modeling gap

Context

From the SNAP research readout follow-ups: congressional-district SNAP results should remain exploratory until independently benchmarked. This issue defines what 'independently benchmarked' means concretely.

Start by reading the existing ACS district benchmark from #240 and PR #241, then inspect the #277 block ladder and #205 vintage translation machinery for reusable crosswalk logic. Confirm how source packages and validation_only families are represented before defining the FNS county-file path. Done means the county-crosswalked benchmark is ingested, compared district-by-district with snap_local_proxy, and remains validation-only until corroboration.