^This^ is the most boring advice ever, and yet its absolutely the right advice.
Anytime you assume things in tech its dangerous, especially so in game dev.
Crucial to your very survival. Advanced Projects often fail just for this reason.
^This^ contains some dangerous assumptions already (from your other thread). Ask anyone whose ever experienced hangs in async level-streaming. 
So you need to prototype and stress test all of your assumptions now by building test-levels asap. Then afterwards, when you have the core setup working, be careful about upgrading or downgrading the engine once again, just to meet some promised new feature or claimed stability. As always upgrading the engine is risky (Russian Roulette suckers game). Unless you have a team of talented C++ coders about, that can fix whatever regression bugs have crept into the engine that release.
Overall, complexity is your enemy here. So if you can get away with anti-complexity or just 1 persistent level, great. If not, if you can get away with just using basic level-streaming, then great as well. But once you start bolting on heavy landscape use or world-partition or manual streaming or a mixture of things, stuff tends to break. So test your assumptions early on in PIE / Standalone / Packaged / Multiplayer now, and again continuously as advised. You can actually have everything from two worlds loaded at the same time (remember to boost paging-file size). Set half as hidden / unticked / no collision. Then customize the lighting setup as needed. However, like everything else in game dev, there’s usually an art to it and gotchas to figure out.