[VL] InsertIntoHadoopFsRelationCommand is used for parquet writing

### Backend

VL (Velox)

### Bug description

I tested the GlutenVelox 1.1 release jar and also built the jar myself with `--enable_hdfs=ON`.
However when running the below query or spark-shell scala code, the `InsertIntoHadoopFsRelationCommand` is falling back to CPU:
```
In spark-shell:
val df=spark.read.parquet("xxx")
df.write.mode("overwrite").format("parquet").save("file:///tmp/yyy")
```
Or
```
In spark-sql:
insert overwrite local directory 'file:///tmp/yyy/' STORED AS parquet select col from db.tbl;
```

My expectation is the parquet write happens on Velox without `VeloxColumnarToRowExec`.

### Spark version

Spark-3.3.1

### Spark configurations

Single node Spark Standalone Cluster with a Spark Master and a Spark worker.

### System information

There is no HDFS cluster. Only a local file system

### Relevant logs

_No response_