Support BINARY in persistent legacy Java UDFs

The BINARY implementation in https://gerrit.cloudera.org/#/c/16066/ supports Java UDFs when the arguments and results are set explicitly, but does not create functions with BINARY types when only the class is given and Impala is supposed to create a function for all overloads of evaluate(),

The reason is backwards compatibility - before BINARY support Impala mapped BytesWritable and BytesArray types to STRING, while Hive maps these types to BINARY. To avoid breaking existing queries in Impala, we cannot create functions with BINARY instead.

If BytesWritable/BytesArray is present as argument, we could create functions with both signatures (on where all these types are mapped to STRING, one where all are mapped to BINARY), but this would not work if these types are only present as return type.