[spark] Fire PaimonMergeInto for data-evolution tables on Spark 4.1 useV2Write=true

Round 8's MergeInto guard unconditionally hands Spark 4.1's RewriteMergeIntoTable
ownership over any SparkTableWithRowLevelOps (useV2Write=true) target. That is
correct for simple V2 row-level tables (row-tracking only), but Spark 4.1's
rule fails the `rewritable` / `aligned` / `needSchemaEvolution` gate for the
more complex data-evolution + blob-field Blob tables. When it silently skips
those plans, the original MergeIntoTable falls through to the physical planner
which throws UNSUPPORTED_FEATURE.TABLE_OPERATION — breaking the 3 Blob merge
tests:
  - Blob: merge-into rejects updating raw-data BLOB column
  - Blob: merge-into updates non-blob column on descriptor blob table
  - Blob: merge-into updates descriptor blob column with external storage end-to-end

Narrow the skip to SparkTableWithRowLevelOps tables that DO NOT have data
evolution enabled. For DE tables Paimon's rule continues to fire and converts
MergeIntoTable to MergeIntoPaimonDataEvolutionTable exactly as Round 3
already did.

Author: kerwin-zk

paimon-spark/paimon-spark-common/src/main/scala/org/apache/paimon/spark/catalyst/analysis/PaimonMergeIntoBase.scala

@@ -48,15 +48,26 @@ trait PaimonMergeIntoBase
     // `AnalysisHelper.allowInvokingTransformsInAnalyzer` so the in-analyzer assertion does not
     // trip.
     val skipV2RowLevelTables = org.apache.spark.SPARK_VERSION >= "4.1"
+
+    // Whether to let Spark's built-in RewriteMergeIntoTable take the plan. On Spark 4.1 we skip our
+    // rewrite when the table already exposes SupportsRowLevelOperations AND data-evolution is
+    // disabled. Data-evolution tables (e.g. Blob tables with 'blob-field' + 'data-evolution.enabled')
+    // cause Spark 4.1's RewriteMergeIntoTable to silently skip (its `rewritable` / `aligned` /
+    // `needSchemaEvolution` gate fails on such tables), leaving the plan as a MergeIntoTable that
+    // the physical planner rejects with UNSUPPORTED_FEATURE.TABLE_OPERATION. Paimon's rule knows
+    // how to route those tables through MergeIntoPaimonDataEvolutionTable, so it must still fire.
+    def spark41V2OwnsPlan(target: LogicalPlan): Boolean = {
+      if (!skipV2RowLevelTables) return false
+      val v2Table = PaimonRelation.getPaimonRelation(target).table
+      v2Table.isInstanceOf[SparkTableWithRowLevelOps] &&
+      !v2Table.asInstanceOf[SparkTable].coreOptions.dataEvolutionEnabled()
+    }
+
     AnalysisHelper.allowInvokingTransformsInAnalyzer {
       plan.transformDown {
         case merge: MergeIntoTable
             if merge.resolved && PaimonRelation.isPaimonTable(merge.targetTable) &&
-              (!skipV2RowLevelTables ||
-                !PaimonRelation
-                  .getPaimonRelation(merge.targetTable)
-                  .table
-                  .isInstanceOf[SparkTableWithRowLevelOps]) =>
+              !spark41V2OwnsPlan(merge.targetTable) =>
           val relation = PaimonRelation.getPaimonRelation(merge.targetTable)
           val v2Table = relation.table.asInstanceOf[SparkTable]
           val dataEvolutionEnabled = v2Table.coreOptions.dataEvolutionEnabled()