[huan233usc]

Writing a batch with an older schema should fill omitted columns with Iceberg
write-default values. This PR adds optional DataWriterOptions::input_schema;
for a required column with initial-default 42 and write-default 7, input
[[1], [2]] writes [[1, 7], [2, 7]].

DataWriter prepares a reusable WriteProjection once in Make and applies it
to each batch. The projection matches fields by ID recursively, preserves
supplied values and explicit nulls, fills missing optional fields with typed
nulls, and materializes missing fields from their write-default. When no
input_schema is provided, or the schemas are identical, the original Arrow
batch is passed through without projection.

Incompatible schemas fail before opening the file, including extra input IDs,
missing required fields without a write-default, optional-to-required fields,
and concrete-to-unknown writes. Existing primitive types must match; read
projection behavior is unchanged.

Parquet and Avro writers remain unchanged and continue to expect batches
matching their complete target schema. The shared projection is implemented in
core using nanoarrow shared buffers without adding an Arrow C++ dependency.

Addresses the write-path portion of #637.

Validation: data_test (184 passed), schema_test (576 passed), avro_test
(238 passed), and parquet_test (110 passed, 1 existing skip). Tests cover two
consecutive batches per format, nested slices and parent nulls, primitive
defaults, explicit null preservation, independent child ownership, and schema
rejection at DataWriter::Make. A separate core/data shared-library build with
bundle disabled passed. clang-format, clang-tidy, git diff --check, and the
Windows CI matrix passed.

[wgtmac]

Is this ready for review?

[huan233usc]

I plan to do some final passes and then will send for review -- thanks

[wgtmac]

After taking a closer look, I'd keep schema alignment and default filling outside the Parquet/Avro writers. They should continue to expect batches matching the full write schema and write the supplied values as-is. Callers can materialize write-default values before invoking the format writer, or DataWriter can handle this through a shared adapter that prepares the projection once and applies it to each batch.

[wgtmac]

Huge thanks to @huan233usc for adding this and completing the v3 default value support! Thanks @shangxinli for the review!