diffrepro: add parallel mode
When diffrepro is executed without a restart script, the queries should be pipelined, similar to orchestrator (instead of waiting for a given chunk to finish, and then restart).
While this mode might be less accurate, it could provide major speedup for CI is used instead of the parallel-restart mode. Currently, diffrepro is a bottleneck: while it takes ~4m for 430k queries in orchestrator, mere 4k differences take almost the same time to verify with diffrepro in parallel-restart mode.