Applied improvement done to the STS unpacker by @fweig to some other unpackers.
Did not do RICH and the TRDs as predicting the upper bound for the digi count is somewhat non-trivial. Probably @a.bercuci and @p.kaehler can comment.
I tested this for a single-threaded calculation and didn't see any effect. My guess is that the gains are system-dependent. I think it might truly shine when parallelization is enabled. I'm guessing that modern CPUs are quite good at predicting / managing their memory layout for single-threaded processes. It certainly cannot hurt to have this in, in any case.
