Optimization of Run-2 TGC-BW Coincidence Window

The storing method of the contents of Coincidence Windows for TGC Big-Wheel is changed for more optimization of memory usage by TGCTriggerDbAlg.
This is one of goal for ATR-21107.

As suggested by the last comment in !38485 (merged), a single std::unordered_map is adopted instead of a load of std::map instances. The unordered_map was faster and smaller memory usage when I compared.

The following test is successfully finished:

RecExCommon_links.sh; athena.py myTopOptions.py

Reco_tf.py --AMI q221 --inputRDOFile /cvmfs/atlas-nightlies.cern.ch/repo/data/data-art/TriggerTest/valid1.410000.PowhegPythiaEvtGen_P2012_ttbar_hdamp172p5_nonallhad.merge.RDO.e4993_s3214_r11315/RDO.17533168._000001.pool.root.1 --maxEvents 100 --athenaopts='--threads=1 --pmon=sdmonfp' --outputRDO_TRIGFile=myRDOTrigger.pool.root --steering=doRDO_TRIG --preExec all:"setMenu='LS2_v1_TriggerValidation_mc_prescale'"

Moreover, I confirmed to output same result of pT determination by the single muon sample.