Benchmark test on CI runs 3 parties linkage

[gusmith]

Threshold 0.8 is chosen for the 3 party case because the results were quite good.

[wilko77]

as discussed, I would prefer to have a sensible threshold in the two party case as well.

[gusmith]

On my machine, results from running the benchmark (remove all the unnecessary values), the provided json may not be valid as I'm removing by hand the values:

{
  'sizes': ['100K', '100K'],
  'threshold': 0.95,
  'groups_results': {
     'accuracy': 0.43175537666706715,
     'false_positives': 0,
     'negatives': 28377,
     'positives': 21561}
  'timings': {'runtime': 27.936643},
},
{
  'sizes': ['100K', '100K'],
  'threshold': 0.8,
  'groups_results': {
    'accuracy': 0.9962553566422364,
    'false_positives': 7289,
    'negatives': 187,
    'positives': 49751
  },
  'timings': {'runtime': 38.349424}
}

So running the 2 parties 100K experiment improves drastically the accuracy using a threshold of 0.8 instead of 0.95, but increases runtime by 10 seconds. So I'll include the change in this PR too.