Re: Support tab completion for upper character inputs in psql

I spent some time contemplating my navel about the concerns I raised
upthread about double-quoted identifiers.  I concluded that the reason
things don't work well in that area is that we're trying to get all the
work done by applying quote_ident() on the backend side and then
ignoring quoting considerations in tab-complete itself.  That sort of
works, but not terribly well.  The currently proposed patch is sticking
a toe into the water of dealing with quoting/downcasing in tab-complete,
but we need to go a lot further.  I propose that we ought to drop the
use of quote_ident() in the tab completion queries altogether, instead
having the backend return names as-is, and doing all the dequoting and
requoting work in tab-complete.

Attached is a very-much-WIP patch along these lines.  I make no
pretense that it's complete; no doubt some of the individual
queries are broken or don't return quite the results we want.
But it seems to act the way I think it should for relation names.

One thing I'm particularly unsure what to do with is the queries
for type names, which want to match against the output of
format_type, which'll already have applied quote_ident.  We can
probably hack something up there, but I ran out of time to mess
with that for today.

Anyway, I wanted to post this just to see what people think of
going in this direction.

			regards, tom lane

PS: I omitted the proposed regression test changes here.
Many of them are not at all portable --- different versions
of readline/libedit will produce different control character
sequences for backspacing, for example.  I got a lot of
failures when I tried to use those tests with this patch;
I've not run down which ones are test portability problems,
which are due to intentional behavior changes in this patch,
and which are due to breakage I've not fixed yet.