hello navv37,
You have already 6 features of SENTINEL2 and another 16 from GLCM. In order for the classification to perform well, the more features we have, the better the classification results. IN your case, you have included way too many features. This will probably overfit your data and your accuracy will drop. It is recommended not to include more than 7-8 features when performing classification. What you can do is to perform a Principal Component Analysis (PCA) on your GLCM results so that to reduce the dimensions down to 2 or 3.
In terms of your first question, you do not have to create training data for both SAR and optical image. Use either the optical or SAR to create training data and use train on vectors because your training data will be in a vector format.
In terms of your second question, all the bands of the SAR and optical data will appear in this box. It depends on you what bands you want to select.
In terms of your third question, the more trees we have, the better the algorithm can capture the complexity of the data achieving higher classification accuracy. On the other hand, the algorithm gets slower the more trees we add