[GH-3383] Preserve GeoParquet XYZ layouts and document M conversion

[jiayuasu]

Did you read the Contributor Guide?

Yes, I have read the Contributor Rules and Contributor Development Guide.

Is this PR related to a ticket?

Closes #3383. The PR title follows the [GH-XXX] format.

What changes were proposed in this PR?

GeoParquet writes currently infer dimensionality from the first coordinate's Z, so declared empty XYZ geometries, NaN Z values, and later finite Z values can become XY. Preserve the supported XY/XYZ layouts with the existing patched WKB writer and derive geometry_types from the emitted WKB header. Reuse one configured writer per geometry-column closure.

Ignore empty envelopes when accumulating the file bbox and omit bbox metadata when no geometry contributes a spatial extent. Empty geometries still contribute their geometry types.

Document the existing measured-input conversion: GeoParquet 1.1 does not support M, so writes project XYM to XY and XYZM to XYZ. The writer remains limited to XY/XYZ with no embedded SRID, retaining the existing WKB encoding flavor. The tutorials advise storing needed M values separately before writing.

The existing JTS limitation for zero-member collections remains: with no coordinate sequence or member, a declared collection layout cannot be retained. Collections with typed empty members are covered by the regression tests.

How was this patch tested?

• Reproduced four failures before the fix: two file-write Z-header failures and two empty-bbox failures; both reader fixtures passed.
• Spark 4.1.3 / Scala 2.13, Java 17: 89 selected GeoParquet tests passed across six suites.
• Spark 3.5.8 / Scala 2.12, Java 17: the same 89 tests passed.
• Both clean Maven reactor builds and Spotless checks passed. git diff --check passed.
• A focused compatibility probe confirmed ordinary two-coordinate point constructors remain XY after geometry serialization.

File tests construct independent ISO WKB fixtures and check both stored WKB and read-back geometries, footer geometry_types, and bbox. Coverage includes empty XY/XYZ primitives, NaN Z, a finite Z after the first coordinate, multipart/nested empty members, both byte orders, null values, measured empty primitives, XYM/XYZM projection, and plain/dictionary input files.

Did this PR include necessary documentation updates?

Yes. Updated the English and Chinese GeoParquet Spark tutorials with XYZ preservation and the explicit M conversion policy.