looking forward improvements

main changes by Daniel:

• triplet seeding search windows are based on UT momentum estimate for high multiplicity events (empirically found to work better for those)
• triplet seeding is done in one kernel, best triplet candidates selected from all triplet searches
• track forwarding to X is also done in one kernel
• precalculated UV windows (per UT track) are used for calculating finer UV windows (per scifi track seed)
• improve throughput of velo search by triplet

main changes by Dorothea: make result deterministic, i.e. tracks -> efficiencies don't depend on execution order

• number of allowed SciFi tracks = # of UT tracks found x factor
• track container accessed through UT track index during x-hit collection
• x filter selects best candidates per UT track based on quality and saves those up to a certain cut-off
• during uv-hit collection, track container is accessed per event
• much less memory is required for the track containers

both changes resulted in lower ghost rate and higher throughput