Switching to the JVM-based libhdfs caused severe performance degradation

### Backend

VL (Velox)

### Bug description

After switching from libhdfs3 to the JVM-based HDFS client, performance degradation was observed in TPC-DS benchmarks for both ORC and Parquet formats. Specifically:

​Parquet on x86: 5% throughput drop
​ORC on x86: 30% throughput drop
​Parquet on ARM: 20% throughput drop
​ORC on ARM: 30% throughput drop."

### Spark version

Spark-3.4.x

### Spark configurations

_No response_

### System information

_No response_

### Relevant logs

```bash

```