When verifying field offset consistency, we shouldn't be checking
base class size when the base class doesn't belong to the current
version bubble. I have fixed this by adding a special case for
the existing fixup encoding indicated by base class size being
zero; please let me know if you find it preferable to create
a new fixup type for this purpose instead. I have also created
a simple regression test that was previously failing when run
against the framework compiled with CG2 assembly by assembly.

Thanks

Tomas