Fix supportVectoredIO for bad hadoop version string

There are cases where the hadoop version info may not be respecting
the semantic versioning. It is the case for the hadoop version
provided in some of the AWS managed services. This causes
a ExceptionInInitializerError while trying to instantiate and ORC
file reader.

Fixing this by defaulting to non vectored IO in case the semantic
versioning is not respected

java/shims/src/java/org/apache/orc/impl/HadoopShims.java

@@ -133,12 +133,18 @@ ByteBuffer readBuffer(int maxLength,
   boolean endVariableLengthBlock(OutputStream output) throws IOException;
 
   default boolean supportVectoredIO() {
-    // HADOOP-18103 is available since Apache Hadoop 3.3.5+
-    String[] versionParts = VersionInfo.getVersion().split("[.]");
-    int major = Integer.parseInt(versionParts[0]);
-    int minor = Integer.parseInt(versionParts[1]);
-    int patch = Integer.parseInt(versionParts[2]);
-    return major == 3 && (minor > 3 || (minor == 3 && patch > 4));
+    try {
+      // HADOOP-18103 is available since Apache Hadoop 3.3.5+
+      String[] versionParts = VersionInfo.getVersion().split("[.]");
+      int major = Integer.parseInt(versionParts[0]);
+      int minor = Integer.parseInt(versionParts[1]);
+      int patch = Integer.parseInt(versionParts[2]);
+      return major == 3 && (minor > 3 || (minor == 3 && patch > 4));
+    } catch (final NumberFormatException e) {
+      // There could be cases where we are not able to parse the version
+      // Here we are defaulting to the non vectoredIO if that is the case
+      return false;
+    }
   }
 
   /**