`fcps|atom` and other datasets with true cardinality <4 show inconsistent results when repeatedly running the same `k`

[imallona]

Datasets with true cardinality < 4 report repeated runs with for some ks, e.g. k, k, k, k+1, k+2 for true k = 2.

Apparently, results for these repeated k are not consistent within these runs.

Thanks Ben for the tip.

[imallona]

this is solved in #30