定製新的RDD類

參照MapPartionsRDD的寫法：

private[spark] class MapPartitionsRDD[U: ClassTag, T: ClassTag](
    var prev: RDD[T],
    f: (TaskContext, Int, Iterator[T]) => Iterator[U],  // (TaskContext, partition index, iterator)
    preservesPartitioning: Boolean = false)
  extends RDD[U](prev) {

  override val partitioner = if (preservesPartitioning) firstParent[T].partitioner else None

  override def getPartitions: Array[Partition] = firstParent[T].partitions

  override def compute(split: Partition, context: TaskContext): Iterator[U] =
    f(context, split.index, firstParent[T].iterator(split, context))

  override def clearDependencies() {
    super.clearDependencies()
    prev = null
  }
}