Does a Dagster job stick to one specific Docker image sha per run? For example if I:
1. push and tag my-job:staging in my container registry
2. update my code location for my-job with my-job:staging and see it refresh
3. start my-job and let it run for a bit
4. while it's still running, push, tag, and update my code location with some changes
5. my-job continues to run and eventually finishes
Does that run of my-job use one consistent Docker image for all the steps in its run? Or does it confuse anything that the definition of my-job:staging (and my-job's code location) change during the run?