Allow prediction in memory, time process

[schilling40]

The project aims to time the process of prediction and segmentation.
To avoid overhead, it allows the prediction to run within memory without saving intermediate results like masks, or intermediate prediction channels.

WIP

I changed the input method for the block shape from nargs=3 to json.loads to be more consistent with other scripts.

[schilling40]

The code is used to segment SGN and IHC crops in memory. This allows us to test and compare the network's performance to that of other segmentation options, such as StarDist, Cellpose, and Micro-Sam.
Additional options were added to the distance U-Net prediction to allow the customization of the boundary distance threshold for the IHC segmentation.