Faster PrVeloUT

Improvements (mostly speed) to PrVeloUT

• Get rid of the tool, do everything directly in PrVeloUT
• Cache look-up table for tolerances locally, instead of accessing the tool for every track. Still need to do the same for the Bdl table.
• Avoid using a std::vector in the clustering stage, as it just creates unnecessary overhead
• Introduce a minimum momentum cut of 1500MeV. More detailed checks needed for possible efficiency loss
• Speed up hit search by precalculating some quantities.
• Don't split hits in top and bottom anymore, gain was not worth the savings in time.
• Also contains a (prototype) version of the method that creates the output tracks, as most of the time is spent in the method that writes out the LHCb::Track in keyed containers. In principle only x,y,z,tx,ty,q/p and the LHCbIDs are needed from VeloUT