added contigName Hive style partitioning to AlignmentRecordRDD

[jpdna]

Will move this PR to bdgenomics once 1018 is merged.
Given this PR, the Parquet directory is laid out with a directory per chromosome (contigName) like

_SUCCESS
_common_metadata
_metadata
_rgdict.avro
_seqdict.avro
contigName=1
    -> part-r-00000-f872ea82-3036-455a-a35d-d043ec386db4.gz.parquet
          ->(in future we will have another layer before the parquet files  )   

            posBin=10000,posBin=20000,...

Later, we will either add a posBin column to Avro or figure out how to allow that column to exist in parquet/dataset but drop from Avro, which will add another layer of directly hierarchy under the 'contigName=N' dirs that bins start pos into 10000 bp bins ( or some other optimal size )

As per discussion in bigdatagenomics#651
such binning should allow a more efficient predicate pushdown of range queries than we may currently get from Parquet.
I'm hoping this strategy is compatible and complementary with the sorted partition mapping system.

The code here can be tested in shell with

import org.bdgenomics.adam.rdd.ADAMContext._
val rdd = sc.loadAlignments("../adam/adam-core/src/test/resources/small.sam")

 val x = rdd.transformDataset(ds => {
     |  import ds.sqlContext.implicits._
     |  val df = ds.toDF()
     |  df.where(df("contigName") === "1")
     |    .as[AlignmentRecordProduct]
     | })

x.saveAsParquet("test_chr_partitioned_parquet")