I'm certainly going to be looking at the problem through a new lens now that I'm aware of these wider issues.
I used to run HA in a VM via qemu/libvirt and it would always fail because when the device reconnects its on a different port and the USB pass through doesn’t work anymore.
Now I’m running the HA VM on proxmox which deals with disconnects reconnects better for pass through devices: it’s able to reconnect it to the vm without my intervention.
Looking at the logs, this happens between a couple times a week to 10+ times a week.
I tried an extension lead (which had no impact, I think this solves zigbee radio issues but not USB issues), using a powered USB hub, using a USB 2 port, a USB 3 port, three different machines... no dice. It really is a bug in the device itself. And Dresden electronics has piss poor support (still waiting for any answer from their email support six months later). The only avenue for “support” is via GitHub issues where other users answer but not actual Dresden electronics employees.
But same experience with Dresden/deCONZ. It's flaky.
Right now I have split networks while I lazily migrate devices away from deCONZ. So far every glitch I've encountered has been on the deCONZ network.
Which is a shame, because I really like the software UI. ZHA doesn't expose as much (or not as simply). But it works better, and that counts for more.