All the transactions generated by the users and those is generates itself the Hive Engine server processes are already stored on the Hive blockchain. If Hive-Engine went down permanently for whatever reason, it would be possible to rebuild its state from those transactions by someone else. (If not, I would be laboring under too optimistic assumptions, which would be bad.)

The centralization pertains to how the data is processed. In contrast, Hive nodes maintain consensus in regard to what actions they take in response to transactions recorded on chain and what Hive nodes do on their own. Hive Engine is just one server.

To decentralize Hive Engine would involve setting up a network of servers that each scan the chain, act upon the transactions and upon generating transactions of their own such as sending second layer token rewards to users - while maintaining consensus on exactly what user-generated transactions have been posted to chain and what Hive-Engine initiated transactions should be posted next.