The true explanation for Peter Geoghegan's trouble report turns out
to be that he has a ~/.inputrc that affects readline's behavior
enough to break this test.  Prevent readline from reading that file.

Also, the best way to prevent TERM from affecting the results seems
to be to unset it altogether, not to set it to "xterm".  The latter
choice licenses readline to emit xterm escape sequences, and there's
a lot of variation in exactly what it will emit.

Revert changes that attempted to account exactly for xterm escape
sequences.  We shouldn't need that with TERM unset, and it was not
looking like a maintainable solution anyway.

Discussion: https://postgr.es/m/23181.1578167938@sss.pgh.pa.us