Implement direct sparse PT

Hybrid KPM allows to apply the sparse Green's function (a.k.a. energy denominators) to a vector. As an alternative, we could use direct sparse solvers. MUMPS implements this using a sparse LU decomposition (ICNTL(24) and ICNTL(25) specifically), and SuiteSparse using a sparse QR.

An advantage of a direct algorithm over KPM is a much more reliable error control. Additionally, once a factorization is computed, applying it is relatively cheap. This would be beneficial for higher orders.