TimeTrackingOutputStream should pass on calls to close() and flush().

Without this fix you get a huge number of open shuffles after running
shuffles.

core/src/main/scala/org/apache/spark/storage/BlockObjectWriter.scala

@@ -101,6 +101,8 @@ class DiskBlockObjectWriter(
     def write(i: Int): Unit = callWithTiming(out.write(i))
     override def write(b: Array[Byte]) = callWithTiming(out.write(b))
     override def write(b: Array[Byte], off: Int, len: Int) = callWithTiming(out.write(b, off, len))
+    override def close() = out.close()
+    override def flush() = out.flush()
   }
 
   private val syncWrites = System.getProperty("spark.shuffle.sync", "false").toBoolean