feat(extract): GISCO extractor + wire all unscheduled extractors

- New gisco.py: proper extractor module replacing scripts/download_gisco_nuts.py.
  Writes uncompressed .geojson (ST_Read can't handle .gz). Fixed partition path
  gisco/2024/01/nuts2_boundaries.geojson; cursor tracking skips re-download monthly.
- all.py: import + register gisco in EXTRACTORS (9 independent, 1 dep)
- pyproject.toml: add extract-gisco entry point
- workflows.toml: add census_usa, census_usa_income, eurostat_city_labels,
  ons_uk, gisco — all monthly, no dependencies
- Delete scripts/download_gisco_nuts.py (superseded)

Unblocks: stg_nuts2_boundaries, stg_regional_income, stg_income_usa,
and 4 downstream models (dim_locations, pseo_city_costs_de,
location_opportunity_profile, pseo_country_overview).

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

extract/padelnomics_extract/pyproject.toml

@@ -21,6 +21,7 @@ extract-census-usa = "padelnomics_extract.census_usa:main"
 extract-census-usa-income = "padelnomics_extract.census_usa_income:main"
 extract-ons-uk = "padelnomics_extract.ons_uk:main"
 extract-geonames = "padelnomics_extract.geonames:main"
+extract-gisco = "padelnomics_extract.gisco:main"
 
 [build-system]
 requires = ["hatchling"]

extract/padelnomics_extract/src/padelnomics_extract/all.py

@@ -7,7 +7,7 @@ A graphlib.TopologicalSorter schedules them: tasks with no unmet dependencies
 run immediately in parallel; each completion may unlock new tasks.
 
 Current dependency graph:
-  - All 8 non-availability extractors have no dependencies (run in parallel)
+  - All 9 non-availability extractors have no dependencies (run in parallel)
   - playtomic_availability depends on playtomic_tenants (starts as soon as
     tenants finishes, even if other extractors are still running)
 """
@@ -26,6 +26,8 @@ from .eurostat_city_labels import EXTRACTOR_NAME as EUROSTAT_CITY_LABELS_NAME
 from .eurostat_city_labels import extract as extract_eurostat_city_labels
 from .geonames import EXTRACTOR_NAME as GEONAMES_NAME
 from .geonames import extract as extract_geonames
+from .gisco import EXTRACTOR_NAME as GISCO_NAME
+from .gisco import extract as extract_gisco
 from .ons_uk import EXTRACTOR_NAME as ONS_UK_NAME
 from .ons_uk import extract as extract_ons_uk
 from .overpass import EXTRACTOR_NAME as OVERPASS_NAME
@@ -50,6 +52,7 @@ EXTRACTORS: dict[str, tuple] = {
     CENSUS_USA_INCOME_NAME:    (extract_census_usa_income, []),
     ONS_UK_NAME:               (extract_ons_uk, []),
     GEONAMES_NAME:             (extract_geonames, []),
+    GISCO_NAME:                (extract_gisco, []),
     TENANTS_NAME:              (extract_tenants, []),
     AVAILABILITY_NAME:         (extract_availability, [TENANTS_NAME]),
 }

extract/padelnomics_extract/src/padelnomics_extract/gisco.py

@@ -0,0 +1,95 @@
+"""GISCO NUTS-2 boundary GeoJSON extractor.
+
+Downloads NUTS-2 boundary polygons from Eurostat GISCO. The file is stored
+uncompressed because DuckDB's ST_Read cannot read gzipped files.
+
+NUTS classification revises approximately every 7 years (current: 2021).
+The partition path is fixed to the revision year, not the run date, making
+the source version explicit. Cursor tracking still uses year_month to avoid
+re-downloading on every monthly run.
+
+Landing: {LANDING_DIR}/gisco/2024/01/nuts2_boundaries.geojson  (~5 MB, uncompressed)
+"""
+
+import sqlite3
+from pathlib import Path
+
+import niquests
+
+from ._shared import HTTP_TIMEOUT_SECONDS, run_extractor, setup_logging
+from .utils import get_last_cursor
+
+logger = setup_logging("padelnomics.extract.gisco")
+
+EXTRACTOR_NAME = "gisco"
+
+# NUTS 2021 revision, 20M scale (1:20,000,000), WGS84 (EPSG:4326), LEVL_2 only.
+# 20M resolution gives simplified polygons that are fast for point-in-polygon
+# matching without sacrificing accuracy at the NUTS-2 boundary level.
+GISCO_URL = (
+    "https://gisco-services.ec.europa.eu/distribution/v2/nuts/geojson/"
+    "NUTS_RG_20M_2021_4326_LEVL_2.geojson"
+)
+
+# Fixed partition: NUTS boundaries are a static reference file, not time-series data.
+# The 2024/01 partition reflects when this NUTS 2021 dataset was first ingested.
+DEST_REL = Path("gisco/2024/01/nuts2_boundaries.geojson")
+
+_GISCO_TIMEOUT_SECONDS = HTTP_TIMEOUT_SECONDS * 4  # ~5 MB; generous for slow upstreams
+
+
+def extract(
+    landing_dir: Path,
+    year_month: str,
+    conn: sqlite3.Connection,
+    session: niquests.Session,
+) -> dict:
+    """Download NUTS-2 GeoJSON. Skips if already run this month or file exists."""
+    last_cursor = get_last_cursor(conn, EXTRACTOR_NAME)
+    if last_cursor == year_month:
+        logger.info("already ran for %s — skipping", year_month)
+        return {"files_written": 0, "files_skipped": 1, "bytes_written": 0}
+
+    dest = landing_dir / DEST_REL
+    if dest.exists():
+        logger.info("file already exists (skipping download): %s", dest)
+        return {
+            "files_written": 0,
+            "files_skipped": 1,
+            "bytes_written": 0,
+            "cursor_value": year_month,
+        }
+
+    dest.parent.mkdir(parents=True, exist_ok=True)
+    logger.info("GET %s", GISCO_URL)
+    resp = session.get(GISCO_URL, timeout=_GISCO_TIMEOUT_SECONDS)
+    resp.raise_for_status()
+
+    content = resp.content
+    assert len(content) > 100_000, (
+        f"GeoJSON too small ({len(content)} bytes) — download may have failed"
+    )
+    assert b'"FeatureCollection"' in content, "Response does not look like GeoJSON"
+
+    # Write uncompressed — ST_Read requires a plain file, not .gz
+    tmp = dest.with_suffix(".geojson.tmp")
+    tmp.write_bytes(content)
+    tmp.rename(dest)
+
+    size_mb = len(content) / 1_000_000
+    logger.info("written %s (%.1f MB)", dest, size_mb)
+
+    return {
+        "files_written": 1,
+        "files_skipped": 0,
+        "bytes_written": len(content),
+        "cursor_value": year_month,
+    }
+
+
+def main() -> None:
+    run_extractor(EXTRACTOR_NAME, extract)
+
+
+if __name__ == "__main__":
+    main()