WIP: split Velo/VeloUT/PV event model for throughput test

Goes with LHCb!1910 (closed) and Brunel!769 (closed). Trying a reduced data exchange format between Velo/VeloUT/PV reconstruction that delays the creation of std::vector<v2::Track> to the end of PrVeloUT.

• Modify PrPixel to output forward (2 states + hits) and backward (1 state + hits) tracks separately
• Modify PatPV3DFuture to accept both these containers
• Modify PrVeloUT to accept the new container type -- as it only sees forward tracks now, the relevant counter changes
• Add ZipFilterIP algorithm to apply the IP cut between Velo/VeloUT
• Add ZipMakeSelection algorithm to produce a selection object if you don't want to apply an IP cut
• Add converter from std::vector<v3::RecVertex> to LHCb::RecVertices
• Add converter from the split velo format to std::vector<v2::Track>
• Update python options, adding converters where needed

Would like to try this in the automatic throughput tests when possible, hopefully lhcb-test-throughput will work. In local tests there are no physics diffs.

Caveats:

• TrackBeamLineVertexFinder does not understand the new velo output format, tests containing it will fail (with an exception in the python configuration)

cc: @pseyfert @sstahl @ibelyaev