Spark execution takes 19 seconds, and it takes 54 minutes after using gluten.

### Backend

VL (Velox)

### Bug description

After using gluten, the performance is greatly reduced, and the execution plan is as follows。
== Physical Plan ==
Execute InsertIntoHiveTable (62)
+- AdaptiveSparkPlan (61)
+- == Final Plan ==
* Project (41)
+- VeloxColumnarToRow (40)
+- ^ ShuffledHashJoinExecTransformer LeftOuter BuildRight (38)
:- ^ ProjectExecTransformer (13)
: +- ^ FilterExecTransformer (12)
: +- ^ WindowExecTransformer (11)
: +- ^ SortExecTransformer (10)
: +- ^ InputIteratorTransformer (9)
: +- AQEShuffleRead (7)
: +- ShuffleQueryStage (6)
: +- ColumnarExchange (5)
: +- VeloxResizeBatches (4)
: +- ^ ProjectExecTransformer (2)
: +- ^ NativeScan hive loan_data_warehouse.ads_did_deviceid_mapping_df (1)
+- ^ RegularHashAggregateExecTransformer (37)
+- ^ InputIteratorTransformer (36)
+- AQEShuffleRead (34)
+- ShuffleQueryStage (33)
+- ColumnarExchange (32)
+- VeloxResizeBatches (31)
+- ^ ProjectExecTransformer (29)
+- ^ FlushableHashAggregateExecTransformer (28)
+- ^ ProjectExecTransformer (27)
+- ^ FilterExecTransformer (26)
+- ^ WindowExecTransformer (25)
+- ^ SortExecTransformer (24)
+- ^ InputIteratorTransformer (23)
+- AQEShuffleRead (21)
+- ShuffleQueryStage (20)
+- ColumnarExchange (19)
+- VeloxResizeBatches (18)
+- ^ ProjectExecTransformer (16)
+- ^ FilterExecTransformer (15)
+- ^ NativeScan hive loan_data_warehouse.ads_did_deviceid_mapping_df (14)
+- == Initial Plan ==
Project (60)
+- SortMergeJoin LeftOuter (59)
:- Project (47)
: +- Filter (46)
: +- Window (45)
: +- Sort (44)
: +- Exchange (43)
: +- Scan hive loan_data_warehouse.ads_did_deviceid_mapping_df (42)
+- SortAggregate (58)
+- Sort (57)
+- Exchange (56)
+- SortAggregate (55)
+- Project (54)
+- Filter (53)
+- Window (52)
+- Sort (51)
+- Exchange (50)
+- Filter (49)
+- Scan hive loan_data_warehouse.ads_did_deviceid_mapping_df (48)

### Gluten version

Gluten-1.3

### Spark version

Spark-3.2.x

### Spark configurations

bin/spark-submit \
 --class org.apache.kyuubi.engine.spark.SparkSQLEngine \
 --conf spark.driver.memory=10g \
 --conf spark.dynamicAllocation.enabled=false \
--conf spark.executor.instances=20 \
 --conf spark.executor.cores=4 \
 --conf spark.executor.extraClassPath=./gluten-velox-bundle-spark3.2_2.12-centos_7_x86_64-1.3.0.jar \
 --conf spark.executor.memory=10g \
 --conf spark.executorEnv.JAVA_HOME=./openjdk-1.8.0.345/java-1.8.0-openjdk-1.8.0.345.b01-1.el7_9.x86_64 \
 --conf spark.files=/home/spark/agent/gluten-velox-bundle-spark3.2_2.12-centos_7_x86_64-1.3.0.jar \
 --conf spark.gluten.sql.debug=true \
 --conf spark.memory.offHeap.enabled=true \
 --conf spark.memory.offHeap.size=10g \
 --conf spark.network.timeout=300s \
 --conf spark.plugins=org.apache.gluten.GlutenPlugin \
 --conf spark.shuffle.manager=org.apache.spark.shuffle.sort.ColumnarShuffleManager \
 --conf spark.yarn.appMasterEnv.JAVA_HOME=./openjdk-1.8.0.345/java-1.8.0-openjdk-1.8.0.345.b01-1.el7_9.x86_64 \
 --conf spark.yarn.dist.archives=/home/spark/agent/java-1.8.0-openjdk.tar.gz#openjdk-1.8.0.345 

### System information

_No response_

### Relevant logs

```bash

```

The report names no source files, query, failing test, or useful logs. Start by reproducing the workload with Spark 3.2.x and Gluten 1.3 using the supplied configuration, then compare execution time and plans with and without Gluten. Done means identifying the regression cause and demonstrating that the workload no longer takes 54 minutes.

Written by the indexing model from the issue text.