feat(data): Phase 2b step 1 — expand stg_regional_income + Census income extractor

- stg_regional_income.sql: accept NUTS-1 (3-char) + NUTS-2 (4-char) codes;
  rename nuts1_code → nuts_code; add nuts_level column; NUTS-2 rows were
  already in the landing zone but discarded by LENGTH(geo_code) = 3
- scripts/download_gisco_nuts.py: one-time download of GISCO NUTS-2 boundary
  GeoJSON (NUTS_RG_20M_2021_4326_LEVL_2.geojson, ~5MB) to landing zone;
  uncompressed because ST_Read cannot read .gz files
- census_usa_income.py: new extractor for ACS B19013_001E state-level median
  household income; follows census_usa.py pattern; 51 states + DC
- all.py + pyproject.toml: register census_usa_income extractor

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

transform/sqlmesh_padelnomics/models/staging/stg_regional_income.sql

@@ -1,15 +1,15 @@
--- Eurostat NUTS-1 regional household income in PPS (dataset: nama_10r_2hhinc).
--- Filters to NUTS-1 codes (exactly 3 characters, e.g. DE1, DE2, …).
--- One row per (nuts1_code, ref_year).
+-- Eurostat NUTS-1 and NUTS-2 regional household income in PPS (dataset: nama_10r_2hhinc).
+-- Accepts NUTS-1 codes (3 characters, e.g. DE1) and NUTS-2 codes (4 characters, e.g. DE21).
+-- One row per (nuts_code, ref_year).
 --
 -- Source: data/landing/eurostat/{year}/{month}/nama_10r_2hhinc.json.gz
--- Format: {"rows": [{"geo_code": "DE1", "ref_year": "2022", "value": 29400}, ...]}
+-- Format: {"rows": [{"geo_code": "DE21", "ref_year": "2022", "value": 31200}, ...]}
 
 MODEL (
   name staging.stg_regional_income,
   kind FULL,
   cron '@daily',
-  grain (nuts1_code, ref_year)
+  grain (nuts_code, ref_year)
 );
 
 WITH source AS (
@@ -34,11 +34,13 @@ SELECT
     WHEN geo_code LIKE 'EL%' THEN 'GR' || SUBSTR(geo_code, 3)
     WHEN geo_code LIKE 'UK%' THEN 'GB' || SUBSTR(geo_code, 3)
     ELSE geo_code
-  END                              AS nuts1_code,
+  END                              AS nuts_code,
+  -- NUTS level: 3-char = NUTS-1, 4-char = NUTS-2
+  LENGTH(geo_code) - 2             AS nuts_level,
   ref_year,
   regional_income_pps,
   extracted_date
 FROM parsed
--- NUTS-1 codes are exactly 3 characters (country 2 + region 1)
-WHERE LENGTH(geo_code) = 3
+-- NUTS-1 (3 chars) and NUTS-2 (4 chars); exclude country codes (2) and NUTS-3 (5)
+WHERE LENGTH(geo_code) IN (3, 4)
   AND regional_income_pps > 0