PV conversion bug fix

Fix for the bug in the PV conversion from Allen to v2 affecting the PV reconstruction efficiency for low-multiplicity PVs.

Vertices reconstructed by pv_beamline_multifitter are asked that the number of selected track for the PV fitting is >=4 (https://gitlab.cern.ch/lhcb/Allen/-/blob/master/device/PV/beamlinePV/src/pv_beamline_multi_fitter.cu#L224). Then, in the conversion step the condition Vertex.nTracks >=4 is repeated (https://gitlab.cern.ch/lhcb/Allen/-/blob/master/Rec/Allen/src/AllenPVsToRecVertexV2.cpp#L66). Since Vertex.nTracks is filled with the sum of the weights of the tracks used for the PV fitting, which is lower than the int number of selected tracks, part of the low multiplicity PVs is not converted, is not considered thus by PrimaryVertexChecker and contributes to the inefficiency.

As shown in the enclosed plots, this MR solves this issue (#135 (closed)) and also contributes to increase the PV efficiency for SMOG2 events, where a higher fraction of PVs show low multiplicity wrt pp.

cc @adudziak