Avro support about rmr2

[piccolbo]

Proof of concept here

https://github.com/RevolutionAnalytics/rmr2/blob/master/pkg/examples/avro.R

Make sure to read about limitations in the comments

[piccolbo]

There is a major issue to fully support formats that need to specify an inputformat (or outputformat), but let's focus on the input side for simplicity. from.dfs calls either streaming with the dumptb command or hdfs dfs -get. Neither accepts input formats as an option, from reading and trying. The former will take a sequence file encapsulated typedbytes data set and write a "pure" typedbytes file, the latter is a copy byte-by-byte. So for now all input formats with a streaming.format components that's not empty or AutoInputFormat (which reads plain text and sequence and is used by dumptb) we can only offer a workaround that is to run a conversion job. Files that are read with from.dfs are in the GB range, so this is workable. Nonetheless we need to review the design, the support for input formats and the implementation options. It's clear that from/to.dfs don't work with formats the way they were expected to.