[BUG] Unnecessary to cache the batches that will be sent to Python in FlatMapGroupInPandas.

[firestarman]

Actually it should be an improvment more than a bug.
The code snip is as below,

        .map { groupBatch =>
          // Cache the input batches for release after writing done.
          queue.add(groupBatch, spillCallback)
          groupBatch
        }

The Python runner will close the batches after writing them to Python, so no need to cache them in the queue for release.
Need to clean this.

[firestarman]

MapInPandas has the similar issue. Code snip