Support native scan for Paimon DSv2 (BatchScanExec) COW tables

[zhuxiangyi]

Background

Auron currently only accelerates Paimon tables that are registered via Hive Metastore and surfaced as HiveTableScanExec. When users configure a Paimon DSv2 catalog (e.g. spark.sql.catalog.paimon = org.apache.paimon.spark.SparkCatalog), the Paimon table is planned as BatchScanExec and native acceleration cannot kick in — the query falls back to Spark row-based execution.

The on-disk layout for the two paths is identical (COW Parquet/ORC files); only the metadata entry point and split planning differ.

Proposal

Add a parallel DSv2 path under thirdparty/auron-paimon that:

1. Detects BatchScanExec whose scan is a Paimon PaimonBaseScan (via reflection, so the module does not hard-depend on a specific paimon-spark-* artifact and stays cross-Spark-version compatible).
2. Validates the splits: COW only — any non-rawConvertible split or any split with deletionFiles causes a fallback to Spark (MOR/MOW is not yet supported by Auron's native scan).
3. Reconstructs partition values from DataSplit.partition() using Paimon's RowDataToObjectArrayConverter.
4. Wraps the planned files in a new NativePaimonV2TableScanExec, mirroring NativeIcebergTableScanExec in structure (metrics, driver-side metric reporting, empty-RDD handling, pb.FileScanExecConf construction).
5. Reuses the existing spark.auron.enable.paimon.scan flag — no new flag.

Scope / Non-goals

• Supports COW only. MOR (primary key, position/equality delete) is intentionally kept as a fallback.
• Hive-registered Paimon path is not touched.
• No predicate / residual filter pushdown to Paimon split planner yet (filters still run at the Spark/Auron operator layer; correctness is fine, but a future improvement can mirror the Iceberg path).
• Statistics (pb.Statistics.getDefaultInstance) are a placeholder for now.

Tests

End-to-end DSv2 integration suite under thirdparty/auron-paimon/src/test/scala/org/apache/auron/paimon/AuronPaimonV2IntegrationSuite.scala covering:

• simple COW select / projection / partitioned table + predicate
• ORC COW table (file.format=orc)
• empty table
• driver metrics (numPartitions, numFiles) via SparkListenerDriverAccumUpdates
• fallback when spark.auron.enable.paimon.scan=false
• fallback for MOR (primary-key) table

PR: #2313