Tell Avro to use String vs. utf8

[kkrugler]

As per Doug Cutting on the Avro list:

The property is "avro.java.string". If the value is "String" then a
java.lang.String is returned. In other words, if the reading schema
is {"type":"string", "avro.java.string":"String"} then String will be
returned instead of Utf8.

This should improve our efficiency. I think we would want to do this when synthesizing an Avro schema (Maven plugin, or from the old API with Fields & types), and probably when reading as we ultimately need to have String objects in the Cascading Tuple.