Add B-physics L2 filter to the mul2IOOvlpRmSequence for the dimuon triggers with 'muCombIO' step (ATR-21003)

1. TrigMultiTrkComboHypo algorithm has been modified to run in a streamer mode. In this mode it makes all possible combinations from objects (tracks or muons) from the appropriate collections and performs vertex fit. If a good combination is found, then it stops running over combinations and copies all decisions with all relevant IDs to the next step. Otherwise the chain execution is stopped. No BLS trigger objects are written out in streamer mode.
2. Trigger chains similar to 'HLT_2mu6_bJpsimumul2io_L12MU6' are configured to use a dedicated close-by muon L2IO algorithm at step2. Now it is equipped with a special StreamerDimuL2ComboHypo which checks the InDet FTF tracks from xAOD::L2CombinedMuon objects can be fitted to the common vertex and its mass falls into the specified mass range. Needed to decrease data rate.
3. Also add a similar StreamerDimuEFComboHypo to the step4 of the Bmumux chains. We check the EFCB muons can be fitted to the common vertex with tight chi2 requirement and narrow mass range before starting the precise InDet tracking in the extended RoIs around those muons.
4. TrigMultiTrkComboHypo has been generalized to work with combinations from N tracks (N = 2, 3, 4, ...). Embedded loops are changed to std::prev_permutation(). Needed to implement trigger HLT_mu20_2mu2btrk_bTauTight for LFV decau tau to 3mu.