The old regex may take quadratic time to scan for potential line
breaks or email addresses starting at every point. Fix it to avoid
scanning from points that would have been in the middle of a previous
scan.
Signed-off-by: Anders Kaseorg <andersk@mit.edu>