Sounds like a plan. Don't forget that HDMI will speed up data rate when connected to a 4K display. Make sure your HDMI cables can handle it. HDMI 2.0 and 2.1 standard cables are ready for that 4K data rate.
Your extractor has that little 5.1 / 2.1 channel switch - always test both modes.
If there is no add-on, you could use a Docker implementation ("nolte" for instance).
Ziggyzhang Don't post your logs as plain text, because it will make this thread hard to read. Use quoting / code formatation for short log snippets, and services like PasteBin for longer logs. Thanks!
Thanks for testing. Next approach: Convert to MKV (by using mkvtoolnix for instance).
...but for some reason it fails with my Samsung TV set: the HDMI signal is not detected properly. As it could be due to "plug&play" aspects of my "smart" TV, I'm looking for other solutions.
I don't think so. If it works without the extractor, it has to work with the extractor, too. That's just a pass-through function. If you still have guarantee, order a replacement.
The USB audio can fail on higher data rate (HD audio formats). So, the extractor solution should be preferred.
I move it to our Hardware sub-forum, because that's the place where people already discuss similar questions.
...so the memory concern is only valid for AI and deep learning users.
No, because AI folks can run it headless, and use the GPU for AI picture detection only.
If you want to run LE at 91MB RAM, good luck!
I'm not a 4K user, but I can give some hints anyway. First read this thread, if you don't want an RPi 4B. I don't suggest the Jetson Nano. If you scroll down that article, you'll see that only 91MB memory are left with graphical display attached. That's terrible.
My RPi is pretty basic, just a power button and status LED. When I find a cool project at the forum, I move it to Showcase. 
Please describe the media formats you use (codec, file format, streaming provider). We're still stepping in the dark here.
PS: Codec Info
Time to try my approach. You're fixed on NTP too much. If you just send the RPi time to Arduino as a loop, it will be more stable in case of network / NTP errors.
Or find the NTP update frequency and set an appropriate delay as first autostart.sh statement. The NTP update and the sync-to-Arduino doesn't have fire at the same time. You can also program a loop with such async timing.
You can activate "Wait for Network" at Settings -> LibreELEC -> Network and put your Arduino script into autostart.sh.
I haven't found related errors at your latest log, so I think it's about a lack of system resources.
Enable HQ Scalers above 50%
Disable HQ Scalers, and hopefully it will set some system resources free.
You're still at log level 0, but that's not your fault. GUI settings overwrite advancedsettings.xml. Try to switch to log level 2 at GUI.
Stop using illegal add-ons, and it might solves your problem.
Because we don't support such add-ons, I have to close this thread.
Navigate to Settings -> System -> Input -> Peripherals -> CEC Adapter and select the appropriate option.
