Note that we use scala.Serializable introduced in Scala 2.9 instead of
java.io.Serializable. Also, case classes inherit from scala.Serializable by
default.
thus clean up the various singletons (RDDCache, MapOutputTracker, etc).
This isn't perfect yet (ideally we shouldn't use singleton objects at
all) but we can fix that later.
Refactored out the agg() and comp() methods from Pregel.run.
Defined an implicit conversion to allow applications that don't use
aggregators to avoid including a null argument for the result of the
aggregator in the compute function.
Note: This test suite currently fails for the same reason that the
Spark Core test suite fails: Spark currently seems to have a bug where
any test after the first one fails.
