feat: Arrow-native geospatial serialization (geospatialAsArrow)

[jatorre]

Summary

Adds support for Arrow-native geospatial serialization via the geospatialAsArrow field (field 6) on TSparkArrowTypes in TExecuteStatementReq.

When enabled, geometry/geography columns are returned as Struct<srid: Int32, wkb: Binary> instead of EWKT strings, enabling efficient binary geometry transfer over Arrow without string parsing overhead.

This follows the pattern established by the existing arrow type options (timestampAsArrow, decimalAsArrow, etc.) and is based on the Databricks implementation of SPARK-54232.

Changes

• internal/cli_service/cli_service.go — Add GeospatialAsArrow as field 6 on TSparkArrowTypes (struct field, getter, IsSet, Read/Write methods)
• internal/config/config.go — Add UseArrowNativeGeospatial bool to ArrowConfig with DeepCopy support
• connection.go — Wire GeospatialAsArrow into TExecuteStatementReq.UseArrowNativeTypes
• connector.go — Add WithArrowNativeGeospatial(enable bool) ConnOption

Usage

connector, err := dbsql.NewConnector(
    dbsql.WithServerHostname(host),
    dbsql.WithHTTPPath(httpPath),
    dbsql.WithAccessToken(token),
    dbsql.WithArrowNativeGeospatial(true), // opt-in
)

Behavior

• Default: false (no change to existing behavior)
• When enabled: Geometry columns arrive as Struct<srid: Int32, wkb: Binary> with WKB binary payload
• When disabled: Geometry columns arrive as VARCHAR containing EWKT strings (current behavior)

Testing

Verified against a Databricks SQL Warehouse:

• Without flag: geometry returned as VARCHAR (POINT(1 2))
• With flag: geometry returned as Struct<srid: Int32, wkb: Binary> with correct WKB bytes

Context

This enables downstream consumers (ADBC drivers, DuckDB adbc_scanner, etc.) to receive geometry data in binary WKB format directly over Arrow, avoiding the overhead of parsing EWKT strings. The Databricks server already supports this via TSparkArrowTypes field 6 — this PR simply exposes the option to Go clients.

[jatorre]

As suggested by @avax94