Minimize allocations on common code paths

There are lots of places in core where a method returns a CompletionStage or CompletableFuture, but it's very common for the stage/future to be completed when the method returns. We can check for that and avoid allocating a lambda and/or a new CompletableFuture.

PersistenceManagerImpl uses RxJava3's Flowable internally to delegate operations to stores, but Flowable adds significant overhead, especially when there is only one store, and the operation only touches one key.