improve resolution of PV reconstruction and fix reconstructed PV nTracks

• reset sum_weights at the beginning of each iteration to get more reasonable number of nTracks
• calculate chi2 w.r.t to the updated vertex position instead of original seed position -> improves resolution, especially for high multiplicity MC PVs
• tweak constants
• align common constants with baseline at https://gitlab.cern.ch/lhcb/Rec/blob/master/Tr/PatPV/src/TrackBeamLineVertexFinderSoA.cpp#L101
• move BeamlinePV constants into separate namespaces
• fix some bugs in cleanup step

To do:

• clean up code and constants
• tweak efficiency and fake rate