Improved regression test driver

I wanted to add a few features and fix a few bugs in the regression test
driver, and I ended up re-writing most of it. I'd like to offer it for
testing.

Features/fixes:

* Use one driver script for both standalone test and test against running
installation.

* Also use only one script for both parallel and serial test schedule,
allow adding of other schedules (e.g., running a lot of simple tests all
in parallel).

* Return useful test summary (x of y tests passed) and exit status, to be
run via `make check' (GNU makefile standards)

* Add option to ignore some failed tests (e.g., "random") for purposes of
the exit status

* Add flag for debug mode

* Avoid use of `make install prefix=xxx', which can (a) yield the test
suite unusable if you overrode specific installation directories,
(b) corrupt your files with wrong hard-coded paths.

* Feable support for running the test suite outside of the source tree,
for binary packages. (incomplete)

What I would like to do is to commit it in parallel with the existing
infrastructure, and if it works out well we can remove the other stuff
before release.

--
Peter Eisentraut peter_e(at)gmx(dot)net http://yi.org/peter-e/