BigTable: row_data _validateChunk performance #4755

Looking at the performance graph for reading a large number (1000) of rows, each row having 10 cells in one column family, the validation of chunks takes a large percentage of time (>25%), as measured by cProfile. Areas for tuning are identified below.

Validation of timestamp_micros ordering
While a row is being read, a validation is made to make sure that cells for one qualifier are in timestamp_micros order (most recent first):

    def _validate_chunk_row_in_progress(self, chunk):
        ...
        previous = self._previous_cell
        _raise_if(self._same_as_previous(chunk) and
                  chunk.timestamp_micros <= previous.timestamp_micros)

    def _same_as_previous(self, chunk):
        previous = self._previous_cell
        return (chunk.row_key == previous.row_key and
                chunk.family_name == previous.family_name and
                chunk.qualifier == previous.qualifier and
                chunk.labels == previous.labels)

This validation may be extraneous because:

1. A RowFilter may have been applied - CellsColumnLimitFilter(1) - which will limit the result to the cell with the most recent timestamp.
2. There may only be one cell value for the qualifier.
3. The BigTable server may ALWAYS be providing the cells in timestamp order.

Two possible tunings:

1. Remove the validation if the condition is unlikely to occur in the ReadRowResponse.
2. If the validation is kept, it could be performed only when qualifier cell lists with length > 1 are detected AFTER the chunks for the qualifier are consumed.