UPLOAD_ERROR causes steps that are unrelated to the bad data to fail

[aiwilliams]

1. Create an uploader for a step (stepId is bound to the uploader)
2. When adding entities/relationships to the graph object store, an "on flush" callback function is passed which pushes the flushing entities/relationships into the queue of the uploader for the step
3. The step that experiences the flush becomes responsible for any failure to upload that chunk of data

This is very confusing during the analysis of any failures.