Wrapper for sparse null space algorithm

This is a feature request for a wrapper that uses the sparse null space finding algorithm in MUMPS.

Finding the null space of large sparse matrices is a common numerical task and (to my knowledge) no python library addresses the issue in a satisfactory way. Below is a rudimentary implementation based on the MUMPS wrapper of kwant.

import numpy as np
import kwant.linalg.mumps as mumps
import scipy.linalg as la

def mumps_nullspace(a):
    a = sp.sparse.coo_matrix(a, dtype=complex)
    a.eliminate_zeros()

    dtype, row, col, data = mumps._make_assembled_from_coo(a, overwrite_a=True)

    mumps_instance = getattr(mumps._mumps, dtype+"mumps")(True)

    n = a.shape[0]
    row = row
    col = col
    data = data

    mumps_instance.set_assembled_matrix(a.shape[0], row, col, data)
    ordering='auto'
    mumps_instance.icntl[7] = mumps.orderings[ordering]
    mumps_instance.job = 1
    mumps_instance.call()

    # Find null pivots
    mumps_instance.icntl[24] = 1
    mumps_instance.job = 2
    mumps_instance.call()

    # Get the size of the null space
    n_null = mumps_instance.infog[28]
    if n_null == 0:
        return np.empty((a.shape[1], 0))
    # Initialize matrix for null space basis
    nullspace = np.zeros((a.shape[1], n_null), dtype=complex, order='F')
    # Set RHS
    mumps_instance.set_dense_rhs(nullspace)
    mumps_instance.job = 3
    # Return all null space basis vectors, overwriting RHS
    mumps_instance.icntl[25] = -1
    mumps_instance.call()
    # Orthonormalize
    nullspace, _ = la.qr(nullspace, mode='economic')

    return nullspace