large regression tests (and benchmarks)

[Stephen Weeks]

> Don't forget that the RPMs are compressed so it might not actually add as much
> space are you guess.

lrtl.sml.gz is still >400K, which is too much for my taste.

> If that isn't acceptable, I would probably vote for
> 2 directories: one for stuff that is in the RPM and one that isn't, but
> maybe one of those can be a subdirectory of the other.  E.g., regression and
> regression/rpm or something.

That sounds good to me.  I'll go for regression and regression/no-rpm.