[WIP][SPARK-56608][PYTHON] Migrate grouped/cogrouped map Arrow UDF verify checks into enforce_schema

[Yicong-Huang]

What changes were proposed in this pull request?

Make ArrowBatchTransformer.enforce_schema the single entry point for Arrow UDF output schema enforcement, and migrate the grouped/cogrouped map Arrow UDF paths (SQL_COGROUPED_MAP_ARROW_UDF, SQL_GROUPED_MAP_ARROW_UDF, SQL_GROUPED_MAP_ARROW_ITER_UDF) to use it instead of the separate verify_arrow_table / verify_arrow_batch helpers plus manual reorder.

enforce_schema is generalized to:

• Accept both pa.RecordBatch and pa.Table.
• Add reorder_by_name: bool = True: name-based matching + reorder + rename (with RESULT_COLUMN_NAMES_MISMATCH) vs positional matching preserving input names.
• Collect all missing/extra/type mismatches before raising (previously raised on first).
• Raise PySparkRuntimeError with existing errorClasses (RESULT_COLUMN_NAMES_MISMATCH / RESULT_COLUMN_TYPES_MISMATCH / RESULT_COLUMN_SCHEMA_MISMATCH) instead of bare-string PySparkTypeError, matching what verify_arrow_result already did.

verify_arrow_table and verify_arrow_batch are deleted; their pa.Table / pa.RecordBatch instance check is inlined at the call site. verify_arrow_result remains only for SQL_ARROW_TABLE_UDF (out of scope for this PR — no benchmark yet).

Why are the changes needed?

Part of SPARK-55388 (Refactor PythonEvalType processing logic). Today output validation is split between verify_arrow_result (friendly errorClass errors) in worker.py and enforce_schema (bare f-string errors) in conversion.py. Consolidating behind enforce_schema gives one code path and one error convention, and drops the redundant "verify + manual reorder" in grouped-map paths.

Does this PR introduce any user-facing change?

Yes, minor: error messages for SQL_ARROW_UDTF (the pre-existing enforce_schema consumer) switch from bare f-strings to the same friendly errorClass-templated format already used by other Arrow UDFs. Error-class names and message formats for grouped/cogrouped map Arrow UDFs are unchanged.

How was this patch tested?

• Existing integration tests in test_arrow_grouped_map.py / test_arrow_cogrouped_map.py already assert the errorClass-templated error format and pass unchanged.
• Unit tests in test_conversion.py updated and extended for the new reorder_by_name, pa.Table input, and count-mismatch paths.
• test_arrow_udtf.py regex updated for the two SQL_ARROW_UDTF error tests.
• ASV benchmarks on CogroupedMapArrowUDFTimeBench, GroupedMapArrowUDFTimeBench, and GroupedMapArrowIterUDFTimeBench (repeat=3) vs upstream master: 52 parameter combinations, 0 regressions at -f 1.05.

Was this patch authored or co-authored using generative AI tooling?

No.