Speed up RowGroup data access methods reducing level of indirection.

RowGroup is the unit of the data sent around CS cluster. It is basically a set of records for fixed size data types + additional storage area for binary data, e.g. strings.
Right now there are lots of RowGroup methods that access RowGroup fixed size data using extra level of inderection. Here is the example:

return *((int64_t*) &data[offsets[colIndex]]);

This expression uses extra assembly instruction to calculate the effective address.
We want to remove offsets[colIndex] part of this and similar expressions in RowGroup code.