Properly account for killed tasks.

The TaskState class's isFinished() method didn't return true for
KILLED tasks, which means some resources are never reclaimed
for tasks that are killed. This also made it inconsistent with the
isFinished() method used by CoarseMesosSchedulerBackend.

core/src/main/scala/spark/TaskState.scala

@@ -26,7 +26,7 @@ private[spark] object TaskState
 
   type TaskState = Value
 
-  def isFinished(state: TaskState) = Seq(FINISHED, FAILED, LOST).contains(state)
+  def isFinished(state: TaskState) = Seq(FINISHED, FAILED, KILLED, LOST).contains(state)
 
   def toMesos(state: TaskState): MesosTaskState = state match {
     case LAUNCHING => MesosTaskState.TASK_STARTING