CoGroupedRDD.

These operations used to wait for all the results to be available in an
array on the driver program before merging them. They now merge values
incrementally as they arrive.

PartitionPruningRDD.

partitions and number of elements in each partition)

package spark to package spark.rdd and updates all references to them.

- SizeEstimator was following through a ClassLoader field of Hadoop
  JobConfs, which referenced the whole interpreter, Scala compiler, etc.
  Chaos ensued, giving an estimated size in the tens of gigabytes.
- Broadcast variables in local mode were only stored as MEMORY_ONLY and
  never made accessible over a server, so they fell out of the cache when
  they were deemed too large and couldn't be reloaded.

The broadcast fix is to store values as MEMORY_ONLY_DESER instead of
MEMORY_ONLY, which will save substantial time on serialization.

Conflicts:

	core/src/test/scala/spark/MesosSchedulerSuite.scala

Conflicts:

	core/src/test/scala/spark/ShuffleSuite.scala

merge results into rather than requiring a new object allocation
for each element merged. Fixes #95.