Reviewing covariance-matrix.c and interaction.c

[Jason Stover]

interaction.c and covariance-matrix.c are ready for further review.

I just committed in interaction.c and covariance-matrix.c that
will correctly compute covariances in one data pass, including data
with categorical variables and interactions, at least for the tests I
have run.

I have run some simple tests. I will start changing glm.q and running
more tests soon.

I have been using this modified copy of glm.q to test:

http://jstover.freeshell.org/pspp/src/language/stats/glm.q

I have been testing it with these syntax files:

http://jstover.freeshell.org/glm.pspp
http://jstover.freeshell.org/glm-a.pspp
http://jstover.freeshell.org/glm-b.pspp

-Jason