PutHiveQL (and PutHive3QL for Hive 3) allows you to execute HiveQL statements without needing a shell or external program to execute. However this use case appears to be more job-oriented (or at least Bulk Synchronous Parallel (BSP) oriented) than flow-oriented, the latter of which is NiFi's focus.
Having said that, there are some processors that can help with this pattern, namely Wait/Notify, but currently they work in a concurrent model, not a parallel one (i.e. the state is stored locally, not for the whole cluster). I don't believe there is currently a mechanism for merging flow files that have been distributed amongst a cluster, but your example wouldn't really need all that except to execute A and B in parallel (vs concurrently, and possibly parallel on different cores, but not guaranteed). If you're willing to give up parallelism, you could split ABC into A,B,C with a FIFO prioritizer into PutHiveQL, then just wait for A,B,C to be executed sequentially. Otherwise I think the solution would get more complicated than you'd want to deal with for this use case. Just my two cents 🙂