Indexer back-to-back jobs after reorg

[whittlbc]

Handle the case where there are multiple rollbacks to the same block close enough together in time to cause multiple index jobs of the same block to run back-to-back, the first not being a replace job, but the second being a replace job. The first job shouldn't even be run at all because the rollback to <= that number has already occurred. Upserts are currently handling everything fine but this could get problematic as more data models are added.