Wrap Parquet and Avro input formats with CombineFileInputFormat

Due to various historical changes in the way Kite works with its own InputFormat, the automatic use of Crunch's CrunchCombineFileInputFormat no longer gets used when reading file-based datasets via Crunch.

This means that each file in an input dataset will result in an additional input split, and therefore an additional map task when reading a dataset. The overhead of a large number of extra map tasks can negatively impact performance.

It would be very useful if Kite were to automatically use CombineFileInputFormat's ability to combine multiple small files into a single input split when processing data via Crunch or MapReduce.