PrForwardTracking redesign
This MR improves the throughput of the PrForwardTracking by currently ~15% tested on a Intel Xeon Silver 4214 (Cascade Lake) and ~5% tested on a reference quanta node (Broadwell). The fast(est) HLT2 configuration in which the Forward Tracking only runs over a subset of SciFi hits also shows ~15% less time spent in this algorithm. The speedup is mainly achieved by introducing a "histogram" to which hits are projected and which makes it easier to spot promising sets of hits to form a track candidate. This also features more static data structures and more vectorization. Also a full cleanup was done. There are now more configurable quantities that allow to tune the trade-off between speed and efficiency. The current working point keeps the efficiencies and ghost rate on the same level as they were before, but reduces the number of clones.
Do not merge before LHCb!2970 (merged)
Goes together with Moore!754 (merged)
Ref updates in Moore!782 (merged) and Brunel!1128 (merged)
Resolves #162 (closed), #157 (closed) and #38 (closed)