feat(reader): handle field ID conflicts in RecordBatchTransformer

[mbutrovich]

Which issue does this PR close?

• Partially address ArrowReader enhancements for Apache DataFusion Comet #1749. Just gonna copy my comments from the test:

    /// This reproduces the scenario from Iceberg Java's TestAddFilesProcedure where:
    /// - Hive-style partitioned Parquet files are imported via add_files procedure
    /// - Parquet files have field IDs: name (1), subdept (2)
    /// - Iceberg schema assigns different field IDs: id (1), name (2), dept (3), subdept (4)
    /// - Partition columns (id, dept) have initial_default values from manifests
    ///
    /// Without proper handling, this would incorrectly:
    /// 1. Try to read partition column "id" (field_id=1) from Parquet field_id=1 ("name")
    /// 2. Read data column "name" (field_id=2) from Parquet field_id=2 ("subdept")
    ///
    /// The fix ensures:
    /// 1. Partition columns with initial_default are ALWAYS read as constants (never from Parquet)
    /// 2. Data columns use name-based mapping when field ID conflicts are detected

What changes are included in this PR?

• Detect conflict in field ID mappings and resolve similar to Iceberg Java BaseParquetReaders.java PartitionUtil.constantsMap()

Are these changes tested?

• New test add_files_partition_columns_with_field_id_conflict
• This fixed 42 tests in Iceberg Java's spark-extensions TestAddFilesProcedure suite when running with Comet's feat: [iceberg] Native scan by serializing FileScanTasks to iceberg-rust datafusion-comet#2528.

[mbutrovich]

Draft while I review some new Iceberg Java failures this created for me.

[mbutrovich]

I think I'll close this in favor of a more comprehensive fix that handles partition specs correctly.