Loosed the tolerance (by 10x) for the matrix_inverse check in

'get_rotation_perms_lists'. The tight tolerance caused failure for
some large supercells when it shouldn't have. This was already committed once but for some reason the symlib latest branch was not included.