Patch field order with avro

[sbarnoud]

What changes were proposed in this pull request?

Fix toBytes() to be not dependent on fields order in the incoming dataset to serialize in Avro. This fix should in the mean time have better performance as we don't need to generate the Avro Schema from the dataset for each row (we only use the user supplied one)

How was this patch tested?

New Unit Test: "avro not dependent on schema field order" This test as been submitted in https://github.com/hortonworks-spark/shc/pull/247 => it fails With this patch => it succeeds

[sbarnoud]

For the remaining problem with the order of the iterator, notice that I'm running Java8 and that your tests seem to be executed in Java7