Fix bias in pv created in Vertex Monitor

Fixes Moore#194 (closed)

The problem is caused by the endvelo covariance not being initialized (because it is unknown), but it shouldn't be an issue as the vertexing algorithms all work with the closestToBeam state. All, except for the TrackVertexer tool that use the TrackStateProvider tool and mix in other random states. This MR fix the Vertex Monitor to explicitely use the closestToBeam state.

Before:

If we remove the endvelo state from the tracks (improve a bit but still a bias):

Force the endvelo covariance to be equal to the closestToBeam covariance (initial solution from @sponce but physically wrong):

Fixed the Vertex Monitor to use only the state closestToBeam:

@sponce @sstahl @rquaglia @sborghi @bimitres