feat: GeoArrow support for bulk ingestion (GEOGRAPHY/GEOMETRY)

[jatorre]

Summary

Adds geospatial column support to the Snowflake ADBC driver's bulk ingestion path. When Arrow columns carry geoarrow.wkb or geoarrow.wkt extension metadata, the driver automatically creates GEOGRAPHY or GEOMETRY columns in Snowflake and converts the data.

• Detects geoarrow columns from ARROW:extension:name field metadata (handles C Data Interface where Go-level extension types are stripped)
• New statement option adbc.snowflake.statement.ingest_geo_type: "geography" (default, WGS84/4326) or "geometry" (any SRID)
• Extracts SRID from geoarrow CRS metadata (PROJJSON or "EPSG:NNNN" format) for GEOMETRY columns
• Unit tests for type mapping and SRID extraction

How it works

1. Detect geoarrow columns from Arrow extension metadata
2. Create table with native GEOGRAPHY/GEOMETRY columns
3. COPY INTO with transform: TO_GEOGRAPHY($1:"geom"::BINARY, true) converts WKB to GEOGRAPHY inline during COPY — no post-processing needed
4. For GEOMETRY columns, SRID is applied via ST_SETSRID if present in geoarrow metadata

Why COPY transform?

Snowflake's COPY INTO from Parquet cannot load WKB directly into GEOGRAPHY/GEOMETRY columns — only CSV and JSON/AVRO support direct geospatial loading from stages (docs).

The initial approach used a post-COPY CTAS pattern (rename → CREATE TABLE AS SELECT with conversion → drop staging). This PR replaces that with a COPY transform that applies TO_GEOGRAPHY/TO_GEOMETRY in the SELECT clause of the COPY subquery, eliminating 3 SQL round-trips and a full table rewrite.

The original CTAS path is preserved as fallback for schemas without geo columns.

Benchmark results

COPY transform vs CTAS approach (50K random points, median of 10 runs):

                       Median      P25      P75      Min      Max   Rows/sec   N
─────────────────────────────────────────────────────────────────────────────────
CTAS (old approach)     8.13s    8.11s    8.26s    7.99s    8.85s      6,150   7
COPY transform (this)   6.11s    5.81s    6.15s    5.34s    7.44s      8,183  10

Speedup (median): 1.33x  (6,150 → 8,183 rows/sec)

The COPY transform also had zero transient failures vs 3/10 for the CTAS path (fewer SQL round-trips = fewer timeout opportunities).

End-to-end with real-world data (Czech Republic OSM Geofabrik):

Dataset	Rows	Throughput	Geometry type
POIs	465,280	38,119 rows/sec	Point
Roads	1,885,651	56,804 rows/sec	LineString
Buildings	5,014,886	68,611 rows/sec	Polygon

At scale (500K rows, single runs):

Dataset	CTAS (old)	COPY transform	Speedup
500K points	29,499 rows/sec	39,339 rows/sec	1.33x
500K polygons	25,189 rows/sec	28,074 rows/sec	1.11x

Export (not in this PR)

Export/read-path geoarrow support is in a separate PR (#100). Detecting GEOGRAPHY/GEOMETRY columns on the read path is non-trivial because:

• With GEOGRAPHY_OUTPUT_FORMAT=EWKB, srcMeta.Type becomes "binary" (type info lost)
• With default GeoJSON format, srcMeta.Type is "object" (same as VARIANT/OBJECT)

Context

This is part of a broader effort to add GeoArrow support across ADBC drivers. Previously opened as apache/arrow-adbc#4114, moved here per maintainer request.

Test plan

• Unit tests for toSnowflakeType with geoarrow extension types
• Unit tests for extractSRIDFromMeta (PROJJSON, simple EPSG string, null, empty, invalid)
• Existing TestIngestBatchedParquetWithFileLimit still passes
• End-to-end tested against real Snowflake with points, lines, and polygons
• Verified GEOGRAPHY column type created in Snowflake via INFORMATION_SCHEMA
• Benchmarked COPY transform vs CTAS approach (10 iterations, median comparison)