Change QueryCompiler get column expressions process from serial to parallel.

When QueryCompiler compile a select sql, the process of getting column expressions is a serial process. The performance is ok when the table is narrow. But when compile a wide table(e.g. 130 columns in my use case), The time-consuming of this step is very high, over 70ms. So I change TupleProjector(PTable projectedTable) from serial for loop to parallel future.

Because this is just modify code performance, not add new feture, so there is no unit test.