SparsePoints silently sorts by species

[max-veit]

From the files of "trying to compute librascal kernels using an existing sparse-point selection" -- the librascal implementation of SparsePoints appears to sort its sparse-point selection by species, without giving the user the slightest hint this is happening. Maybe there's a good (?) reason for it in the code logic, but it's highly unexpected (and undocumented!) behaviour if I supply a sparse point selection that is not sorted by species.

For ease of comparison with other codes, I would like to have a version of SparsePoints that either (1) doesn't sort by species or (2) undoes the sorting after computing the kernels. Solution (2) might be the easier one.