[SPARK-16350][SQL] Fix support for incremental planning in wirteStream.foreach()

## What changes were proposed in this pull request?

There are cases where `complete` output mode does not output updated aggregated value; for details please refer to [SPARK-16350](https://issues.apache.org/jira/browse/SPARK-16350).

The cause is that, as we do `data.as[T].foreachPartition { iter => ... }` in `ForeachSink.addBatch()`, `foreachPartition()` does not support incremental planning for now.

This patches makes `foreachPartition()` support incremental planning in `ForeachSink`, by making a special version of `Dataset` with its `rdd()` method supporting incremental planning.

## How was this patch tested?

Added a unit test which failed before the change

Author: Liwei Lin <lwlin7@gmail.com>

Closes #14030 from lw-lin/fix-foreach-complete.