Implementing this interface indicates to Pig that a given loader
can be used for MergeJoin. It does not mean the data itself is ordered,
but rather that the splits returned by the underlying InputFormat
can be ordered to match the order of the data they are loading. For
example, files splits have a natural order (that of the file they are
from) while splits of RDBMS does not (since tables have no inherent order).
The position as represented by the
WritableComparable object is stored in the index created by
a MergeJoin sampling MapReduce job to get an ordered sequence of splits.
It is necessary to read splits in order during a merge join to assure
data is being read according to the sort order.