StructBatchReader should always skip processing on the rootReader

As of right now the processing in StructBatchReader is as follows:

• `nextBatch` method : Perform directly on the children with the assumption that the rootReader has no nulls.
• `seek`, `skipRows`, `startStripe` methods: Operate on the rootReader

Propose that the StructBatchReader be made consistently to operate on the children and never on the rootReader.