Refactoring: add small batch processing

[dzenanz]

Currently, analysis_pipeline function takes a single image as input. Change this function, or add an additional function, to allow processing multiple images in a tight loop. This should allow more efficient use of GPU models (segmentation and registration), by loading them into GPU, invoking them on multiple images, then unloading them. The rest of the processing can then be run in parallel on the CPU for each image.