Make it easier to reliably test data observers

In my experience, testing data observers is somewhat brittle. Even with explicit transactions I sometimes get "delayed" notifications that have been triggered by the previous operation.

Let's take this test (ObjectBox 2.2.0 + RxJava 2):

boxStore.runInTx {
    boxStore.boxFor(SomeEntity::class.java).put(SomeEntity())
}

val testObserver = TestObserver.create<Optional<SomeEntity>>()

boxStore.runInTx {
    repository.find(123).subscribeWith(testObserver)
}

testObserver.awaitCount(1)
testObserver.assertValueCount(1)

Sometimes, I get a value count of 2 from the original put operation. For completeness, the code of the find() method:

fun find(id: Long): Observable<Optional<SomeEntity>> {
    return Observable.create { emitter ->
        val query = this.boxStore.boxFor(SomeEntity::class)
            .query()
            .equal(SomeEntity_.id, id)
            .build()

        query.subscribe().observer { result ->
            emitter.onNext(result.firstOrNull().toOptional())
        }
    }
}

Assuming I did not misunderstand the docs or misuse the ObjectBox API, it would be great if ObjectBox could offer some method e.g. to block until all notifications triggered by the last transaction have been issued.

Start with the runInTx, TestObserver, and query.subscribe().observer sequence shown in the issue, then review the ObjectBox data-observer and transaction behavior it exercises. Done means a reliable way to wait for notifications from the preceding transaction, demonstrated by a test that consistently receives the expected single value.