Use "long int" internal calls for UMFPACK

Using this, we get rid of the memory limitations for larger matrices. As it turns out it does not affect run times at all.

Discussion is found here: #33 (closed)