Spurious ALLEGRO failure

[Robert Goldman]

I'm not sure what the upshot of this is.

Currently, the tests test images and other bundles by starting up a
subsidiary lisp, and then checking its console output.

Writing a dribble file wrapper around that will be pretty painful.

For now I propose to continue to use build.exe and buildi.exe for
Windows, pending a re-think of how the lisp-invocation library (which we
use for testing) should handle console output on windows.

As Faré points out, our tests are single-threaded, so this seems
adequate for now.

If I'm missing something, please LMK.

Otherwise, I will probably press on to release 3.1.6.

P.S. I don't have time to do a whole lot of debugging of the SBCL +
run-program issue right now, although I have been able to replicate it.