Reduce PV fake rate

The PV fake rate is rather high (~6%). Investigate the cause and how to reduce it (without losing efficiency). One possible reason is the reconstruction of a single true PV as two separate ones. Investigate cut against this, like here https://gitlab.cern.ch/lhcb/Rec/blob/master/Tr/PatPV/src/PVOfflineTool.cpp#L292

Another thing to look at is that some fake PVs seem to have NaN entries in their covariance matrix.

• investigate cut against duplicates
• investigate fake PVs with NaN for z uncertainty