prefer MsgCounter over Warning(...) and Error(...)

In order to avoid reference file updates due to printout differences like

-- Reference file
+++ Actual output
@@ -31,5 +31,5 @@
    FunTupleBase_Particles/B02Dpi_Tuple  #=80      Sum=80          Eff=|( 100.0000 +- 0.00000 )%|
-Run2SSProtonTagger_521e5c97.Tagg... SUCCESS  #WARNINGS   = 1        Message = 'No convergency has been reached'
-Run2SSProtonTagger_521e5c97.Tagg... SUCCESS  #WARNINGS   = 1        Message = '_iterate_opt: No convergency has been reached'
-Run2SSProtonTagger_521e5c97.Tagg... SUCCESS  #WARNINGS   = 1        Message = 'fit(): failure from _iterate()'
+Run2SSProtonTagger_86eda44e.Tagg... SUCCESS  #WARNINGS   = 1        Message = 'No convergency has been reached'
+Run2SSProtonTagger_86eda44e.Tagg... SUCCESS  #WARNINGS   = 1        Message = '_iterate_opt: No convergency has been reached'
+Run2SSProtonTagger_86eda44e.Tagg... SUCCESS  #WARNINGS   = 1        Message = 'fit(): failure from _iterate()'
 RFileCnv                               INFO dumping contents of /NTUPLES/FILE1
…

(the above is from test_davinci_tupling_with_flavour_tagging) and replace them with counter updates instead (which recognize algorithm renaming due to hash changes), and as a side-effect to make the above code thread safe, by replacing the use of Error(...) and Warning(...) with Gaudi::Accumulators::MsgCounter<...>.