RFR: 8239357: Revert gcc implementation of offset_of

[Kim Barrett]

Please review this change to undo the (small) portion of JDK-8238281
[1] related to the HotSpot offset_of macro for gcc/clang.  That change
should not have been made.  See CR for details.

I did not reinstate the redefinition of offsetof; we shouldn't be
using that macro, and having the warning enabled will catch
questionable cases.  (I also considered but didn’t poison it.)

CR:
https://bugs.openjdk.java.net/browse/JDK-8239357

[1] https://bugs.openjdk.java.net/browse/JDK-8238281

Webrev:
https://cr.openjdk.java.net/~kbarrett/8239357/open.00/

Testing:
mach5 tier1

With a temporary change to use offsetof with a non-standard-layout class,
verified the expected build failures occur with gcc and clang.