cleanup_hardware has registers live across it when called from the
boot sector code; it must preserve all registers.

Remove unused variable BufSafeSec; it was identical to BufSafe in
every way.

At the moment, BufSafe* are always constants; consider if they can be
swapped out for constants or if we want to retain them for flexibility.

Add support for sha256crypt and sha512crypt to the library.  Not yet
included in the menu system, need to decide if it's worth 10K or if
the code can be realistically compacted.

For 32-bit code, 32-bit operations are shorter than 16-bit operations,
so use them.

Create a guard section .uibss that explicitly is covering the
potential partial-sector-clobber zone, and contains items which are
safe to assume will only be used after full image load.

Clean up the transition to PM for shuffle-and-boot-to-PM, and set LDTR
and TR to keep Intel VT happy.

The image size loaded is bigger than the real image; this is due to
the final sector being fractional.  Move variables needed during that
stage from .bss to .bss1.

XXX: Need to audit to make sure this doesn't affect other derivatives
and/or other variables, too.

The 386 and 486 requires a jmp after any setting of cr0.  It can be a
near jmp, but those processors do not correctly synchronize without
it.

We have jmps after setting cr0 in all spots except one, so make that
happen correctly.

Clobber "mov cs,ax" with "nop nop" even though that adds a byte to the
source; it's a lot easier to comprehend while debugging.