Implement and test sorted neighbor list

[singraber]

In #77 it turned out to be beneficial to introduce a sorted neighbor list (large cutoff required for real-space part of Ewald summation).

Such a sorted neighbor list could be useful also for SF calculation when different cutoff radii are used. Maybe it would help to create a per-atom map of "neighbor cutoffs" corresponding to different cutoff radii, e.g.

struct Atom
{
    map<double, size_t> neighborCutoffs;
}

Then, during SF setup we initialize the required cutoffs:

neighborCutoffs[sf->getRc()] = 0;

When the neighbor list is sorted, we have to loop over it and determine the neighbor cutoffs for all radii:

// Prepare list of desired cutoffs.
vector<double> cutoffs;
for (auto const& nc : neighborCutoffs) cutoffs.push_back(nc.first);

// Loop over neighbors.
size_t ic = 0;
for (size_t in = 0; in < neighbors.size(); ++in)
{
    Neighbor const& n = neighbors.at(in);
    // If neighbor distance is higher than current "desired cutoff" then add neighbor cutoff.
    // Attention: a step in the neighbor list could jump over multiple cutoffs => need while loop.
    while (n.d > cutoffs.at(ic) && ic < cutoffs.size())
    {
        neighborCutoffs.at(cutoffs.at(ic)) = in;
        ic++;
    }
    if (ic >= cutoffs.size()) break; // All neighbor cutoffs already found.
}

Finally, in the SFs calculate() function we have to replace size_t numNeighbors = atom.numNeighbors; by

size_t numNeighbors = atom.neighborCutoffs.at(rc);

Note: This will not be very useful for SFs with identical cutoff radius (traditional BP), but could increase speed for polynomial SFs.