[MLton] bin/regression

[Stephen Weeks]

> 
> My reading of the bin/regression script is that when a regression program 
> fails to compile, the script should print:
> 
> "compilation of FOO failed with -type-check true"
> 
> and then more on to the next regression program.
> 
> However, for me, it appears to simply abort the regression script.

That happens because of the "set -e" at the top of the script.  I
think it's good to have there by default, to get a blunt notice that
things have gone wrong.  But it is useful to comment out for some
testing.