feat: add dictionary_columns to scan API for memory-efficient string reads

[tanmayrauth]

Exposes dictionary_columns: tuple[str, ...] | None = None on Table.scan() and DataScan, threading it through to PyArrow's ParquetFileFormat so that named columns are read as DictionaryArray instead of plain large_utf8. This dramatically reduces memory usage for high-cardinality repeated JSON/string columns (issue #3168) and addresses the general scan parameter extensibility request (issue #3170).

Key implementation details:

• ORC files are guarded — dictionary_columns is only passed for Parquet
• ArrowScan.to_table() rebuilds the Arrow schema with dict types before the empty-table fast-path so schema is consistent regardless of row count
• DataScan.to_arrow_batch_reader() rebuilds target_schema with dict types to prevent .cast() from silently decoding DictionaryArray back to plain string
• DataScan.__init__ declares and stores the param so TableScan.update() (which uses inspect.signature) preserves it across scan copies

Fixes #3168, closes #3170

Rationale for this change

Are these changes tested? Yes

Are there any user-facing changes? No

[tanmayrauth]

@kevinjqliu @Fokko can you please review and approve this?

[tanmayrauth]

@geruh @kevinjqliu @Fokko can you please review this implementation?

[github-actions]

This pull request has been marked as stale due to 30 days of inactivity. It will be closed in 1 week if no further activity occurs. If you think that's incorrect or this pull request requires a review, please simply write any comment. If closed, you can revive the PR at any time and @mention a reviewer or discuss it on the dev@iceberg.apache.org list. Thank you for your contributions.

[github-actions]

This pull request has been closed due to lack of activity. This is not a judgement on the merit of the PR in any way. It is just a way of keeping the PR queue manageable. If you think that is incorrect, or the pull request requires review, you can revive the PR at any time.