rS313948 partially fixed --mmap behavior but took an odd and incomplete approach. This commit
generally reverts it and does it the more correct way- by just consuming the rest of the buffer
and moving on.
The previous approach was wrong because that's just now how --mmap operates. The part of the loop
now behind a filebehave != FILE_MMAP is only to see if we hit an EOL/EOF in part of the file we had
not yet read in. There's no point in doing this when for --mmap because we already have the entire buffer
mapped, so if we hit this loop then we're really looking at EOF. The previous fix was resulting in getting
hit by a (SIG)BUS in certain scenarios, such as reported in PR 219402, because we were copying past the end of
the lnbuf for reasons that don't make sense.
PR: 219402