[SPARK-14078] Streaming Parquet Based FileSink

This PR adds a new `Sink` implementation that writes out Parquet files.  In order to correctly handle partial failures while maintaining exactly once semantics, the files for each batch are written out to a unique directory and then atomically appended to a metadata log.  When a parquet based `DataSource` is initialized for reading, we first check for this log directory and use it instead of file listing when present.

Unit tests are added, as well as a stress test that checks the answer after non-deterministic injected failures.

Author: Michael Armbrust <michael@databricks.com>

Closes #11897 from marmbrus/fileSink.