Make decoding of BX map more efficient

Profiling has shown that the decoding of the BX map uses around half the time of the processing. One clearly sees this in attached graphs: CountBX decodes the BX map, the sort function seems to be used too in that process and within the FillOrbit[Muon|Calo] functions there are some noticeable accesses to the vector carrying the BX info too. We should be able to optimise all that.

Flame graph for the muon processor: