Fix yet another "how did this ever work?" thing in genesis.
* Fixups on x86-64 would read 8 bytes, 4 of which were total junk,
coming from the next machine instruction. Little-endianness,
and the fact that all (SETF BVREF-n) setters ignore bits,
made this work essentially by accident.
* The test for whether to record in %code-fixups a relative fixup
for 32-bit was wrong but got the right answer. It meant to compare
the absolute address in question against the code component address
range, but instead compared the computed displacement.