Change gptldr from relocating 0xfff1 bytes of boot2 to relocating 0x20000
bytes of boot2. Since we're in 16-bit mode, we can't copy all 128kB at
once; instead we loop four times and copy 32 kB each time.

This change was made necessary by an upcoming increase in the size of the
boot2 binary; should it increase further, the COPY_BLKS value can be
adjusted without anyone needing to remember 8086 assembly language again.

Requested by: allanjude
Tested by: allanjude
MFC after: 1 week