Misuse column name instead of column index during data reader's serialization

[sparkliu]

During the serialization of data reader, most of code uses column name instead of column index, for example, get column type, check null, get data. Actually as the final purpose is to return the whole dataset, there is no such situation that a column exists in dataset but not serialized, also the sequence of property is exactly the same with the sequence of column. So the serialization can be optimized to use column index to reduce column searching.

[sparkliu]

Fix patch