Re: A really subtle lexer bug

Here's what I will push unless there's something important I missed.

I split the regression tests between create_operator.sql and
polymorphism.sql, because => is really "named argument syntax" rather
than an operator as such, and polymorphism.sql is where the existing
tests were for that.

I tried to keep the three lexers as closely matched as possible, even
though psqlscan.l doesn't actually need some of the changes.

-- 
Andrew (irc:RhodiumToad)