It's unclear what the actual issue is (debug logs demonstrating the problem would help) but disabling DRMPRIME is a workaround and not a fix. You will probably/eventually find that other media will play badly because you've disabled the optimised decoding path and are forcing media to be less efficiently handled.
Any idea?
LE 9.0 uses the legacy vendor kernel and amcodec in Kodi. LE 10.0 onwards uses the upstream kernel which reimplements hardware decoding (but the implementation is incomplete) and Kodi dropped support for amcodec in 2017 as part of an intentional purge of prioprietary codecs and refocus on standards.
TL/DR: the codebases are completely different and while most media plays reasonably well on a C2 it is not pefect and will not work with all media for all users. The upstream codebase is mature and stable for practically everything except media capabilities, where work stalled years ago and has never resumed. You may find that some media is badly encoded and re-ripping from the original sources or re-encoding makes things more playable.
NB: mkv is a container not a format so it's hard to comment further without knowing what the actual format was (although HEVC is the most likely answer).
You'll need to share a full debug log for anyone to comment, log snippets are pointless.
Once we see the full log it'll probably reveal that you're trying to play a 4K 10-bit HEVC file and that will fail because there is no HEVC hardware decoding support on S905X3 (SM1) devices and the CPU is too weak for software decoding.
Changes will be reset on HDMI hotplug events, and that includes reboot.
What LE device are you playing the videos on?
The current workaround is a script (re)configuring the DRM layer from userspace. If someone can come up with a kernel patch to avoid hacking things from userspace: a) we'll know what area of kernel driver code to focus Intel devs on, b) i'll accept/merge a temporary patch to see what issues arise: the patch probably needs to cap bits-per-pixel and those calculations are used for lots of things so there is potential for unintended fallout and the 'simple' change might not be simple.
LE 12.2 and 13.0 have dropped the legacy nvidia driver, and the number of ‘newer’ cards in active use is tiny (we have been telling people not to use nvidia cards for a decade; people have listened) so we may simply drop nvidia support completely and have a more simple life. It will suck to do that, but as nvidia continues to either ignore standards or implement them with some catch that eliminates LE use, so be it.
There's an obvious kernel splat in the log, but nothing obvious about why that occurred, and Kodi has started in 'off-screen' mode.
Are you using any kind of HDMI adapters or an Argon case?
I'm wondering if the audio glitch is similar to the N100 issues here: https://github.com/LibreELEC/LibreELEC.tv/pull/8588 - the script in that PR is not going to be merged as it's too much of a hack, but you could copy/paste it locally and experiment. It basically finds the active DRM connector and limits the max bits-per-pixel to 10-bit as 12-bit seems to cause issues. The root cause is unknown but I have a hunch it's related to bandwidth and UHD video + HD audio maxes out something resulting in audio dropouts.
You might want to add something like video=HDMI-A-1:1920x1080M@60 to kernel boot params in syslinux.cfg in case (for reasons unknown) the box is trying to use a problematic 4K mode. This will clamp initial kernel DRM output to 1080@60 to stop that. There's no impact on Kodi switching to 4K during playback later.
There's no shortcut/script for it, but the backup process we use in the LE settings add-on is just a 'tar' command. If the box is only hosting Tvheadend you only need to stop Kodi and then tar the contents of /storage/.kodi/userdata/addon_data and any other dirs that you've used in your TVH configuration as the rest of the OS and Kodi should be using default settings.
NB: If you are thinking to update to LE 12.2 (or LE13 nightlies) Tvheadend 4.2 is no longer supported and only Tvheadend 4.3 exists in the 12.2/13.0 addon repo you are updating to.
In theory de-featuring the board to a single HDMI connector shouldn't be an issue, but if you fuzzed one thing it's possible other things were damaged too. Historically i'm able to play 4K30 VP9 content from YouTube, although I didn't test since the rather niche regression below was possibly added (and can't test from current location).
There is one software regression we are triaging, noted when playing 4K H264 on an RPi5; much higher resolution than 720p VP9 but also software decoding so there's a chance your issue could be the same thing. Log in to the RPi5 over SSH, run rpi-eeprom-config -e and add SDRAM_BANKLOW=3 to the end of the file, then reboot. Any better after? NB: If no difference, it's not the same issue and you can revert the change.
Usually there is no autoupdate first week or two,
There will never be auto-update from 12.0.x to 12.2.x.
LE only auto-updates within the same release series, e.g. 12.0.0 > 12.0.1 > 12.0.2. It never auto-updates between major releases like 12.0.x > 12.2.0 or 12.x > 13.x as major releases often contain breaking hardware changes and/or Kodi major version changes.
For example. forced update from 12.0.2 > 12.2.0 would be an issue for RPi5 users running Tvheadend 4.2 (doesn't exist in 12.2) or an x86_64 device with older nVidia card would rebot to no Kodi home screen (GPU drivers don't exist in 12.2).
You can manual update in the settings add-on if you've read the release notes and nothing impacts you.
The laptop has a proper WiFi antenna allowing 'max' transmission rates and so much single-core CPU capacity that single vs. multi-threading makes no practical difference. The CPU design is general purpose and the fully built system uses a large amount of power and generates a lot of heat even when handling trivial tasks.
The RPi5 has a rubbish antenna so WiFi will always be at a disadvantage when compared to Ethernet, and the ARM CPU single-core performance is much lower so single vs. multi-threaded has a more noticeable impact. The SoC package is a lot more efficient and a fully built system uses considerably less power and generates less heat for a trivial workload.
As usual, the moral of the story is .. use Ethernet 
Random guess is that pget is multi-threaded and get is not.
Best to continue the discussion in the Tvheadend forum thread: https://tvheadend.org/d/9200-server-spec/
