Fix RMSE computation in 'ErrorCrit_RMSE'

Currently the denominator uses only count the number of values in the observed times series.
Anyway the model can return other NA values due to streamflow transformations (i.e. "inv" and "log").

So we have to replace:

Denom <- sum(!is.na(EC$VarObs))

by:

Denom <- sum(!is.na(EC$VarObs) & !is.na(EC$VarSim))