[VL] Flaky test on CSV on Spark-4.1

### Backend

VL (Velox)

### Bug description

https://github.com/apache/gluten/actions/runs/23478921533/job/68373808924?pr=11802

```
corrupted ZSTD compressed csv respects ignoreCorruptFiles *** FAILED ***
2026-03-24T14:57:01.5354862Z null did not equal "FAILED_READ_FILE.NO_HINT" (SparkFunSuite.scala:359)
2026-03-24T14:57:01.5695665Z 14:57:01.568 WARN org.apache.spark.sql.execution.datasources.csv.GlutenCSVv1Suite:
```

```
2026-03-24T14:57:44.6465291Z - corrupted ZSTD compressed csv respects ignoreCorruptFiles *** FAILED ***
2026-03-24T14:57:44.6466213Z null did not equal "FAILED_READ_FILE.NO_HINT" (SparkFunSuite.scala:359)
2026-03-24T14:57:44.6728926Z 14:57:44.672 WARN org.apache.spark.sql.execution.datasources.csv.GlutenCSVLegacyTimeParserSuite:
```

### Gluten version

_No response_

### Spark version

None

### Spark configurations

_No response_

### System information

_No response_

### Relevant logs

```bash

```

Start with the failing cases named in GlutenCSVv1Suite and GlutenCSVLegacyTimeParserSuite, using the linked GitHub Actions run and the assertion location in SparkFunSuite.scala:359. Reproduce the corrupted ZSTD CSV failure and determine why the expected "FAILED_READ_FILE.NO_HINT" value is sometimes null; done means the affected test cases pass consistently.

Written by the indexing model from the issue text.