DRILL-1389: Incorrect results when reading nullable data out of parquet
Fixed numerous issues in the nullable reader for parquet, removed some cruft from previous attempts to fix earlier issues that did not properly address the root causes. The reader overall is simpler and cleaner. Verified against a variety of existing a new files with complete result checks comparing against Steven's parquet reader that uses the higher level interface.
Wrote some new test tooling to allow for comparisons of large result sets.
Re-enabled the fixed binary support that was accidentally removed in some of the last changes before 0.5.
Fixed a bug where we were initializing the column data reader to read the uncompressed size of the column chunk, rather than the on-disk size.