Remove duplicate LICENSE and NOTICE files from benchmark JARs

Similar to https://github.com/apache/iceberg/issues/17406

**These local JMH artifacts are not published, so releases are unaffected**. Update their shade configuration to retain only the intended `META-INF/LICENSE` and `META-INF/NOTICE`, while preserving required attribution in those canonical files.

The shaded benchmark uber-JARs contain multiple LICENSE/NOTICE variants:
- `parquet-benchmarks.jar`
- `parquet-plugins-benchmarks.jar`

## Reproduction

```bash
# Build all default and vector-plugin JARs without running tests.
./mvnw --batch-mode -Pvector-plugins -DskipTests clean package

# Check all final JARs; exclude only Maven Shade backup artifacts.
(
failed=0
while IFS= read -r -d '' jar; do
licenses=$(jar tf "$jar" | grep -Eic '(^|/)LICENSE([._-].*)?$' || true)
notices=$(jar tf "$jar" | grep -Eic '(^|/)NOTICE([._-].*)?$' || true)

if [ "$licenses" -eq 1 ] && [ "$notices" -eq 1 ]; then
echo "PASS: $jar"
else
echo "FAIL: $jar (LICENSE=$licenses, NOTICE=$notices)"
jar tf "$jar" | grep -Ei '(^|/)(LICENSE|NOTICE)([._-].*)?$'
failed=1
fi
done < <(
# original-* JARs are temporary backups created by Maven Shade.
find . -type f -path '*/target/*.jar' \
! -name 'original-*.jar' \
-print0 | sort -z
)
exit "$failed"
)
```

Observed failures:

```text
FAIL: ./parquet-benchmarks/target/parquet-benchmarks.jar (LICENSE=9, NOTICE=4)
FAIL: ./parquet-plugins/parquet-plugins-benchmarks/target/parquet-plugins-benchmarks.jar (LICENSE=4, NOTICE=3)
```

Expected: every final JAR contains exactly one LICENSE and one NOTICE file under `META-INF/`.

Inspecciona la configuración de shade en los módulos parquet-benchmarks y parquet-plugins/parquet-plugins-benchmarks. Ejecuta ./mvnw --batch-mode -Pvector-plugins -DskipTests clean package y, después, inspecciona los JAR finales de benchmark con las comprobaciones proporcionadas. Se considera completado cuando cada JAR final contiene exactamente un META-INF/LICENSE y un META-INF/NOTICE, conservando las atribuciones requeridas.

Escrito por el modelo de indexación a partir del texto del issue.