[training] Remove spatial dims selection. Display to user and update UI when we know the spatial dims of their dataset

It does not make sense that the user can select the number of spatial dims in their image- just one additional error they can make when running training. It would be good to get the number of dims from their input images, and display to the user instead.

- When user selects input images for training, get dimensions along with number of channels
- display this information to the user
- disable `z_patch_size` edit if user has 2d images, enable for 3d.