[Java] Avro to Arrow converter should reuse the same VectorSchemaRoot

When originally reviewing the code I suggested a new VectorSchemaRoot on each call to the iterator.  After further discussions on the mailing list, it seems that this is an anit-pattern for working with VectorSchemaRoot, we should update the code to update a single VectorSchemaRoot.

After this change it should be easier to use Avro converter with other components of the library (i.e. filewriter) which also make use of a single VectorSchemaRoot.

**Reporter**: [Micah Kornfield](https://issues.apache.org/jira/browse/ARROW-9613) / @emkornfield

_{**Note**: *This issue was originally created as [ARROW-9613](https://issues.apache.org/jira/browse/ARROW-9613). Please see the [migration documentation](https://github.com/apache/arrow/issues/14542) for further details.*}

Start by locating the Java Avro converter and its iterator, then read how VectorSchemaRoot is created and returned on each call. The change is done when the converter updates and reuses one VectorSchemaRoot, allowing it to work with components such as the filewriter; run the converter’s existing tests if available.

Written by the indexing model from the issue text.