New scheme for managing regress test result files

Since we have to go through the process of regenerating regress test
result files anyway, now seemed like a good time to take care of
something that's been bugging me for a while. I have just committed
changes that allow multiple platforms to share platform-specific
regress test result files.

For example, there are a lot of machines where the int2 regress test
produces
ERROR: pg_atoi: error reading "100000": Result too large
instead of the reference platform's
ERROR: pg_atoi: error reading "100000": Numerical result out of range
We can now have all these platforms share a single result file,
which I've called expected/int2-too-large.out, rather than having
to have duplicate result files for each such platform. There is
a mapping file src/test/regress/resultmap that identifies which file
to use for each platform --- it's a lot like src/template/.similar,
if you've messed around with that.

So far I've only put entries into resultmap for my own platform (HPUX)
but I'm sure many more will get added over the next few weeks.

The parallel regress test script, run_check.sh, doesn't seem to work
with this scheme yet. It *ought* to work but, at least on my machine,
it seems like /bin/sh has problems with nested "while read" loops.
I've run out of steam to work on this for tonight --- perhaps someone
else can see how to fix it.

Also, I updated src/test/regress/README but didn't touch the SGML
doco yet...

regards, tom lane