feat: apply write defaults in Parquet and Avro writers

[huan233usc]

Summary

Writing an Arrow batch with an older schema currently requires callers to materialize every column added by schema evolution. This PR lets Parquet and Avro writers align batches to the target schema and populate omitted columns with Iceberg write-default values.

For a newly added required column with initial-default = 42 and write-default = 7:

input schema:  required int id
write schema:  required int id, required int added
input rows:    [[1], [2]]
stored rows:   [[1, 7], [2, 7]]

The file is written after added exists, so the omitted column uses write-default, while initial-default remains a read-time rule for older files. This addresses the write-path portion of #637 and includes the work in #941.

• Add optional input_schema to DataWriterOptions and WriterOptions. Omitting it defaults to the target schema. The Writer::Write interface stays unchanged; both writers always call the shared alignment helper, which forwards the original Arrow C Data batch unchanged when schemas match.
• Reuse Project with write-default selection and disabled type promotion. Missing optional fields become null, missing required fields without a write-default fail, and explicitly supplied nulls stay null.
• Align struct, list, and map children recursively by field ID. Preserve existing array buffers, offsets, and validity through ArrayData; materialize unknown inputs as null arrays of the target type. Reject unknown input field IDs at every nesting level.

Review pointers

Start with MissingColumnUsesWriteDefault and UnknownInputWritesTypedNulls in test/default_value_test.cc, which write and read through the public DataWriter for Parquet and both Avro backends. AlignBatchForWrite in arrow/arrow_c_data_util.cc handles Arrow materialization; ProjectionOptions in schema_util.h selects the read/write policy.

The alignment regressions cover forwarding identical schemas without import/export or changes to C Data ownership, existing collections when a scalar default is added, defaults and field reordering inside nested collections, sliced structs with parent nulls, unknown field IDs inside collections, and typed nulls. The new alignment tests fail against the previous implementation and pass with these changes.

Validation

Built all five test targets and ran their complete test binaries:

Suite	Passed	Skipped
data_test	193	0
schema_test	566	0
avro_test	238	0
parquet_test	110	1
manifest_test	200	18

All suites completed with zero failures. Skips cover codec availability and unsupported format/version cases. clang-format, clang-tidy on the five changed implementation files, and git diff --check passed.

[wgtmac]

Is this ready for review?

[huan233usc]

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