WIP: Improve speed of PV fitting

• add grid y-dimension to PV multi-fitter
• blocks in grid y-dimension loop over vertex seeds in each event
• parallelize loop over tracks using threads -> one block takes care of fitting one vertex seed. In one fit the loop over all tracks is parallelized using threads. Shared memory is used to save the result of each thread, which is then reduced

TO DO:

• optimize number of threads and blocks
• investigate where speed-ups can be gained (fewer atomicAdd, syncthread, etc.)
• investigate non-deterministic behaviour