Refactor data reader classes for vectorized reads

[samarthjain]

This refactoring takes care of providing a method to customize behavior of how identity partition columns are handled in RowDataReader and BatchDataReader.

The bulk of the change is to move the piece of code on how identity partition columns are handled in BaseDataReader to RowDataReader. Currently, BatchDataReader isn't used when identity partition columns are projected #838.

[prodeezy]

Since this is part of the Base Readers api, would be good to add a docstring on what this is supposed to do and where it's being used.

[rdblue]

@samarthjain, can you describe these changes in the description and give more context about why they are needed?

[samarthjain]

@rdblue, @prodeezy - I have updated the PR describing the changes. Also pushed a commit to add doc for the method.

[prodeezy]

lgtm @samarthjain

[rdblue]

#1004 was merged, so I don't think we need this refactor any longer.

There were only 2 cases where we need to project before returning a batch or row to Spark: if extra columns were projected for filtering (fixed by #1004) or when using a JoinedRow to add identity partition values. Since vectorized reads don't support adding identity-partitioned values, there are no longer any cases where we need to project.

Also, when vectorized reads do support identity-partitioned values, we should be able to use the idToConstant map that Avro and Parquet currently use. Then we would create columns in the right order to begin with and won't need a projection at all.

Since this isn't needed and I already talked with @samarthjain, I'm going to close it.