Training for adafs and mvfs

[TopoKunst]

First of all thank you for this foundational work. I've got some questions for the training of AdaFs and MvFs.

Both two models set its optimizer method as "normal", while in train_one_epoch function of fs_trainer, it seems that the feature selection part of these two models would not be trained? I find only lpfs, optfs and models with optimizer method of darts would be trained as in the training function.
I would appreciate it if you could help me with my questions. Thank you!