DiskAccess::read_exact_into did not work correctly if current_offset
wasn't aligned to 512 bytes.

For example, if self.current_offset is 3 and len is 1024:
- end_addr will be 3+1024=1027
- start_lba will be 3/512=0
- end_lba (1027-1)=2
read_exact_into would then read 3 (!) sectors with lbas 0, 1, and 2
even though the buffer can only hold 2 sectors (1024 bytes).

To fix this, only allow seeking to the start of a sector.
DiskAccess::read_exact_into works correctly if the offset is a multiple
of the sector size.

There were few uses of DiskAccess::seek that didn't seek to the start
of a sector. All those uses then use DiskAccess::read_exact to read
data at the offset. Unlike DiskAccess::read_exact_into,
DiskAccess::read_exact can already handle non-aligned offsets. To fix
the bad seek calls, we turn read_exact into a combined seek+offset
function that works for non-aligned offsets.