Optimise probes filter for small limits

[alexey-yarmosh]

80% of measurement requests today use limit: 1, we can optimise for that.

On every request today:

1. Filter all probes.
2. Shuffle all.
3. Group by asn+city.
4. Build a list by picking a probe from every group until all groups exhaust.
5. Pick the required number of probes from the beginning. Just first probe in most of the cases.

Instead:

On every probes list update:

1. Build probe groups by asn+city.
2. Sort groups desc by size.

On every request:

1. Filter groups by checking its first probe (since probes are the same in one group).
2. Pick the required number of probes from groups using random round-robin pick.

I'm oversimplifying here, but the idea looks valid for me.

[MartinKolarik]

I considered this but would rather not touch this area at the moment as the selection is rather complex and the gains somewhat questionable.