The alignment required for the second union (has 64-bit size) causes
a hole between the first and second union. Move the remaining data
there.
In 64-bit build, shrinks brw_reg from 24 bytes to 16 bytes. And by
consequence, shirnks fs_inst from 200 bytes to 160 bytes, making it
use one less cacheline.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30822>