Support Iceberg _spec_id metadata column in native scan

[weimingdiit]

Is your feature request related to a problem? Please describe.

Currently, native Iceberg scan only supports the _file metadata column. Queries that project other Iceberg metadata columns fall back to Spark, even when the metadata column is file-level and can be materialized as a constant per scanned data file.

For example, _spec_id is a file-level Iceberg metadata column. It represents the partition spec ID of the data file containing the row. Unlike _pos, it does not require row-level materialization, so it can be supported in the same way as _file.

Related code:

• IcebergScanSupport.isSupportedMetadataColumn only allows MetadataColumns.FILE_PATH.
• NativeIcebergTableScanExec.metadataPartitionValues only materializes _file.

Describe the solution you'd like

Add native Iceberg scan support for the _spec_id metadata column.

The implementation can treat _spec_id as a per-file partition value:

1. Mark MetadataColumns.SPEC_ID as a supported metadata column in IcebergScanSupport.
2. Build a mapping from data file path to FileScanTask.file().specId().
3. Extend NativeIcebergTableScanExec.metadataPartitionValues to materialize _spec_id as an integer literal.
4. Add integration tests for queries such as:
   • select _spec_id from iceberg_table
   • select id, _file, _spec_id from iceberg_table

The native scan should continue to fall back for row-level metadata columns such as _pos.

Additional context

Iceberg defines _spec_id as a required integer metadata column. Since the value is constant for all rows in a data file, it can be passed through the existing native scan partition-value mechanism already used for _file.

This is a focused native Iceberg scan coverage improvement and should not require an AIP.