yes, I would say that. A RF works like a classification tree, it splits your input data using thresholds so it can be used to describe your target variable. It doesn’t matter if one of them is Intensity and another is coherence.
Please see here: random forest - Do I need to normalize (or scale) data for randomForest (R package)? - Stack Overflow (answer of Hong Ooi) It is discussed there that stretching is even harmful for RF applictions.
Conversion is in the Raster menu:
Please see the corresponding help section for the different options.
After conversion, note the new range:
If you have trouble with NaN areas you have to work with the NoData Value in the band properties. This user has had similar problems: Value error: Water boundary with the image background is gone!