Re: GiST support for UUIDs

I wrote:
> I'm kind of inclined to change uuid_parts_distance to just convert
> a given pg_uuid_t to "double" and then apply penalty_num(), as is
> done in gbt_macad_penalty.

Pushed with that change and some other mostly-cosmetic tweaking.

One not too cosmetic fix was that gbt_uuid_union was declared with the
wrong return type.  That's probably mostly harmless given that core GiST
pays little attention to the declared signatures of the support functions,
but it's not a good thing.  This would've been caught if anyone had
thought to run the amvalidate functions on the updated extension.
I think I will go and put a call to that into the regression tests of
all the contrib modules that define new opclasses.
        regards, tom lane