Optimize CaloFuturePedestal

Some issues have to be fixed in the CaloFuturePedestal algorithm introduced in !2952 (merged) Should be solved by !3107

• use of new thread-safe histograms

• fix repetition of bin filling

• RMS plot should be handled in online's AutomaticAnalysis framework instead of the consumer's finalize unless the new sink does it (to be decided)

• Fill only in empty-empty (?)