As of now MATLAB does not support setting different 'gamma' values for separate folds in the same training scenario. What I can suggest is, you can manually separate data into different folds and train different cross-validation models using 'fitcdiscr'. This may require some understanding about how the prediction algorithm actually works.
To know more about prediction follow the link below.
Prediction Using Discriminant Analysis Models - https://www.mathworks.com/help/stats/prediction-using-discriminant-analysis-models.html
