feat(csharp): define geospatial Arrow types

[cathleeny]

What's Changed

• Recognize Reyden's internal GEOMETRY and GEOGRAPHY transport as struct<srid:int32,wkb:binary>.
• Define the public GeoArrow WKB field contract using Arrow Binary, the geoarrow.wkb extension name, and CRS/edge metadata.
• Parse default, fixed, and ANY SRIDs and validate malformed or conflicting transport tags.
• Add focused SQL type and transport-schema tests.

The Databricks tagged struct is an internal server transport contract. PR #700 converts it before exposing results to ADBC consumers.

Stack: 1 of 3.

Testing

• Stack-wide C# unit suite: 981 passed.
• CI is running on the rewritten stack.

[peco-review-bot]

Verdict: 1 Medium · 1 Low

Looks solid overall — clean canonical-type definition with good parse/validation test coverage. One medium concern: the GEOMETRY/GEOGRAPHY → StructType mapping is wired into the live SEA schema path (StatementExecutionStatement.cs:712) with no serializing stream rewriting geospatial batches to that struct, so it relies on Reyden already emitting struct<srid,wkb> natively — worth confirming given the PR defers client representation to a later stack. One low note on TryGetTag throwing despite the Try* name. Nit (summary-only): the SRID regex -?\\d+ accepts a literal GEOMETRY(-1), which maps to the same tag as the ANY sentinel (srid = -1) — benign today since real SRIDs are non-negative, but the two become indistinguishable on round-trip.

[lidavidm]

This doesn't seem to use GeoArrow types? Shouldn't the driver use the standard types instead of defining its own keys?

[peco-review-bot]

Verdict: 2 Low

Looks good — a clean, well-tested new file defining the geospatial Arrow type contract; 2 low-severity notes only. The parsing, tag validation, SRID defaults (GEOMETRY=0 / GEOGRAPHY=4326 / ANY=-1), and GeoArrow metadata generation are internally consistent and match the Databricks geometry(0) base-name convention noted in the learning log. The notes concern the Try-prefixed method that throws and a missing test for the documented untagged-struct-returns-false path.

[peco-review-bot]

✅ No issues identified by the review bot.

[eric-wang-1990]

For GEOMETRY(ANY) the comment says rows carry EWKB, but the field is still labeled geoarrow.wkb. My understanding is the GeoArrow spec expects ISO WKB here, so a strict reader might reject the EWKB SRID flag or silently drop the SRID. Can you confirm against the spec what readers (e.g. geoarrow-c, GDAL, geopandas) do with EWKB in a geoarrow.wkb column? If it isn't supported, options include ISO WKB plus a separate per-row SRID, or documenting ANY as a known deviation.

[cathleeny]

GeoArrow recommends ISO WKB where possible, but allows consumers to accept EWKB. For a fixed SRID geometry, we use ISO WKB. But for GEOMETRY(ANY), ISO WKB cannot preserve each row’s different SRID, so EWKB is the only lossless single column representation.

geoarrow-c and geopandas accept EWKB, but GDAL rejects it (this is only for GEOGRAPHY(ANY))

[peco-review-bot]

✅ No issues identified by the review bot.