QR decomposition with column pivots has been added to the latest SVN
code. Regular QR decomposition breaks down when a matrix is singular,
but with column pivots it can finish the entire decomposition. Useful
when trying to solve systems which might be singular and is much
faster than SVD.
The current code base will probably become 0.18 next week.