WIP: change prpixelalg output

Changes to perform a throughput tests with smaller event model objects. In hindsight, this is largely the same work as in

!1530 (closed)

but probably it is not nearly as complete.

In summary:

• introduce new class VeloPix::Hit, VeloPix::Track. The track is not owner of the hits: hits will be put in a separate container in the svt store.
• make a PrPixelAlgorithm implementation that uses the new objects.
• make a PrVeloUT algorithm that uses them (and outputs a VeloUT track + UTHit following the same model.)
• introduce a version of TrackBeamLineVertexFinder algorithm that's implemented on the input track type, both for VeloPix::Track and the original v2::Track.

Besides this, some small fixes/other 'improvements':

• introduced a correlation element in the hit error matrix, though admittedly we'll probably never use it in the PrPixel track fit. (we should use it in the full fit, though)
• put the Kalman filter that is currently in PrPixel in a public place (PrKernel/PrPixelKalmanFit) such that it can be used in the 'refit' that Laurent has implemented
• changes the output of TrackBeamLineVertexFinder from a v2::RecVertex to a PrimaryVertex . The latter can do very efficient refitting at little costs.

The new algorithms are all introduced with the suffix New, which is horrendously ugly, but allows me to run them in parallel. This should certainly be fixed/removed before merging (if ever).