FIX: Accuracy Issue Due to Pseudo Inverse

[skailasa]

When solving the linear system for the equivalent densities from the check potential we use a pseudo inverse, however this is a source of error due to the pseudo inverse which comes from multiplying its factors together as the diagonal matrix of singular values has very small/large numbers. This is alleviated by multiplying the diagonal matrix of singular values by either the left or right singular vectors, and storing the pseudo inverse in two components. This is known as the backward-stable pseudo inverse, though i haven't listed all of the details here, they can be found in the original paper of PVFMM. This fixes accuracies obtained in the FMM, and we are able to get all the digits we expect for higher orders.