Reloc table opcodes are used to filter out coincidental TVector-like
things.
This problem was identified while trying to debug all the XTOC glue that
didn't make sense. We also ignore this glue.
This time I was swapping the meaning of the RelocBySectC/D opcodes. This wasn't terribly bad, because they were rarely used. I identified the bug while figuring out how to handle the rare fragment that uses XTOC glue to access its own guts in a patchable manner.
RelocSmRepeat/RelocLgRepeat opcodes were repeating the *next* opcode, not the
*previous* opcode. This completely wrecked the relocation parsing for a small
number of fragments. I left in the debug improvements I made while debugging.