Document the boundary vintage of districts/*.h5 files

[DTrim99]

Problem

The HF Hub dataset policyengine/policyengine-us-data/districts/{XX-NN}.h5
doesn't declare which Congress's boundaries were used to allocate
households to each district. Users can't tell whether pairing the h5
with a particular geojson (118th, 119th, 120th) is internally
consistent.

Concrete failure mode: a dashboard renders a 119th-Congress
choropleth on top of impact numbers sampled against 118th boundaries
(or vice versa). District labels match, but the underlying populations
don't.

Asks

• Add a boundary_year / congress_session attribute to each h5
  output (e.g., "congress": 119, "tiger_year": 2024,
  "genz_vintage": "GENZ2024").
• Document the build script's geographic input source in the
  dataset card (Census GENZ / TIGER vintage, plus the Hugging Face
  release tag that produced the current h5 set).
• Backfill the attribute on the existing artifacts if practical;
  otherwise mark them as unknown and document going forward.

Why this matters now

We're staring down a 119th → 120th transition that touches 8 states
(see #1144). Without boundary-vintage metadata, dashboards will
silently mix vintages during the rollout window.

Related

• Plan the 119th → 120th boundary transition (8 states) #1144 — 119th → 120th boundary transition (8 states)
• Move the canonical congressional-district geojson into this repo #1145 — move the canonical geojson into this repo
• Audit 118th → 119th boundary catch-up (GA, NC, NY, AL, LA) #1146 — 118th → 119th boundary catch-up audit