[SPARK-16697][ML][MLLIB] improve LDA submitMiniBatch method to avoid redundant RDD computation

## What changes were proposed in this pull request?

In `LDAOptimizer.submitMiniBatch`, do persist on `stats: RDD[(BDM[Double], List[BDV[Double]])]`
and also move the place of unpersisting `expElogbetaBc` broadcast variable,
to avoid the `expElogbetaBc` broadcast variable to be unpersisted too early,
and update previous `expElogbetaBc.unpersist()` into `expElogbetaBc.destroy(false)`

## How was this patch tested?

Existing test.

Author: WeichenXu <WeichenXu123@outlook.com>

Closes #14335 from WeichenXu123/improve_LDA.

mllib/src/main/scala/org/apache/spark/mllib/clustering/LDAOptimizer.scala

@@ -28,6 +28,7 @@ import org.apache.spark.graphx._
 import org.apache.spark.mllib.impl.PeriodicGraphCheckpointer
 import org.apache.spark.mllib.linalg.{DenseVector, Matrices, SparseVector, Vector, Vectors}
 import org.apache.spark.rdd.RDD
+import org.apache.spark.storage.StorageLevel
 
 /**
  * :: DeveloperApi ::
@@ -472,12 +473,13 @@ final class OnlineLDAOptimizer extends LDAOptimizer {
         gammaPart = gammad :: gammaPart
       }
       Iterator((stat, gammaPart))
-    }
+    }.persist(StorageLevel.MEMORY_AND_DISK)
     val statsSum: BDM[Double] = stats.map(_._1).treeAggregate(BDM.zeros[Double](k, vocabSize))(
       _ += _, _ += _)
-    expElogbetaBc.unpersist()
     val gammat: BDM[Double] = breeze.linalg.DenseMatrix.vertcat(
       stats.map(_._2).flatMap(list => list).collect().map(_.toDenseMatrix): _*)
+    stats.unpersist()
+    expElogbetaBc.destroy(false)
     val batchResult = statsSum :* expElogbeta.t
 
     // Note that this is an optimization to avoid batch.count