It is very likely to be swapping things out that then never get swapped back in. In which case there's no harm being done, and a slight good in that there's more RAM for caching.
There is the issue of how the VM is stored. If it is on, say, lvm-thin, then the storage can handle the snapshot and the VM is in raw format. If it is on a file storage like NFS, then the image itself must do it, for example by being in qcow...
No, it still doesn't make sense. Proxmox is somehow supposed to know if this was an "intended" shutdown rather than a VM crash, and do something different in that case. And you want "on demand" machines to have high availability? What the heck...
What is the point of HA if you're going to be shutting the VM down for extended periods and it should not run on a cluster restart? On the face of it this does not make any sense.
The stuff about setting a slightly higher voltage and/or lower frequency in BIOS seems relevant though. It might also pay to look at what c-states are enabled and whether you have the AMD microcode installed.
Other than that I got nuthin'.
It is because he's trying to sync to a Windows DC and per usual Windows does things different. See for example:
https://support.scc.suse.com/s/kb/Synchronize-chrony-with-a-Windows-NTP-Server?language=en_US