ALL distributed BFT-consensus systems are based on synchronicity assumption.

That's wrong, there are BFT algorithms without synchronicity assumptions:

• Haveras Hashgraph is asynchronous
• Honeybadger

Those are two recent ones, in the literature you'll find more.

Now what does it mean to be "synchronized"?

Synchronization in the context I was talking about is making timing assumptions.

Making timing assumptions means that we can assume that a message from one server to another will have a "maximum" roundtrip.
Asynchronous protocols don't make any timing assumptions in this regard.

Now, if we know there is a maximum time, then we can ALWAYS detect if a faulty and thus, we can avoid a lot of the problems and can achieve safety and liveness with a lower percentage (51%).

Going back to DPoS. Yes, every block needs N^2 to be finalized. But if we add more witnesses this will spread out over a longer period of time as well so that will only increase the latency and not be a problem for throughput at all.