Depending on as-yet-incompletely-explained factors, readline/libedit
might choose to emit screen-control escape sequences as part of
repainting the display.  I'd tried to make the test patterns avoid
matching parts of the output that are likely to contain such, but
it seems that there's really no way around matching them explicitly
in some places, unless we want to just give up testing some behaviors
such as display of alternatives.

Per report from Peter Geoghegan.

Discussion: https://postgr.es/m/CAH2-WznPzfWHu8PQwv1Qjpf4wQVPaaWpoO5NunFz9zsYKB4uJA@mail.gmail.com