Regression: Fix unnecessary reprocessing of all images

[mihow]

Summary

When a new job is created, all images are determined to need reprocessing, instead of only those remaining in the collection. This is a problem because if a job fails, it has to start from the beginning.

This seems to have been introduced in #684

How to Reproduce the Bug

• Create a collection with a few images (10 random)
• Process the collection with a pipeline
• Process them again (hit Retry)
• Look at the "filter count" in the job logs, it will be 10 out of 10 again.