PrLongLivedTracking speedups

Some changes to increase the speed of PrLongLivedTracking.

• exploit existing sorting of hits.
• write a dedicated fit function without outlier removal
• reduce calling the distance function
• write a fast distance function for small z differences.
• Use float throughout PrDownTrack, i.e. also for the magnet position and the track vector.

Goal is not to change the performance numbers efficiency and ghost rate. Should increase the speed of PrLongLivedTracking by ~15 - 20%.

Evaluation on 2025 data shows very small differences in the spectra and Ks and Lambda peaks (red is new, black is old).

@mveghel @dovombru @vsvintoz

Some more optimizations are possible, but given this should help with HLT2 throughput, I think it makes sense to get this merged already now.