Begin to clear up the mysteries of why the restriction doesn't work.

Everything's still a mess, but at least the tests are passing. No more
do we check for exact equality between K_W and K_star_W_star; instead
we just check that many properties agree. In the paper we have an
isomorphism, and equality holds. However, my isomorphism isn't an
isomorphism when using a coordinate system -- and we have to use a
non-normal one in Sage because we can't normalize vectors over
QQ. Huh.