Simplify thresholing UX and workflow

**Please describe the problem that needs to be solved by this feature**
Currently, we allow thresholding on images that are the result of prediction, a folder of images, or any image dropped into the napari viewer. This makes things necessarily complex.

I propose that users should only be able to apply thresholds on images that are the result of prediction. We should not have to select a input directory or drag and drop any images into the viewer.

Following this UI/UX change- we can simplify code as follows

- update ThresholdingModel once prediction finishes with the results, storing paths of all segmentations
- use that list of images as inputs to thresholding, so no need to re-select images before starting
- No need to rely on `source_path` metadata or `layer.source.path` to get image paths

**How important is this feature to you? (Software team will re-prioritize all tickets based on current workload)**
medium

**Describe the solution you'd like**
First we need to run this UX change idea through the UI team. The above solution is my idea on how to simplify things, but we should agree on a good solution that makes this ux easy for users and more manageable for the debs.