Handle manual rebuilds correctly

[prolativ]

If one tries to rebuild a failed job (e.g. because formerly the executor ran out of resources), after the build finishes (with a success or failure) downstream jobs are triggered for rebuilding. Currently, even if they succeeded before they will fail on publishing artifacts during the rebuild because the artifacts repository doesn't allow reuploads. What should be done in such cases? E.g.

1. Don't rebuild downstream jobs if they were built successfully before - but then we won't know whether using the newly build artifacts from the upstream job would change anything
2. Allow reauploading artifacts - should we then invalidate the results of all transitively downstream jobs and stop the ones that are currently running?