From my experience on Z2M, you should see the ‘neighbor that is not part of the network’ message permanently go away after removing and re adding the the device that has the orphaned device entry. As I said, getting the network to a state where there are none of these messages seems to reduce the occurrences of device dropping of the network for a period. However, not eliminating the issue.
To your points, Zigbee gives users almost no visibility to the movement of packets around it’s network. For example, you can command a bulb to turn on or off and between executions of this command you can see a human noticeable difference in how long between ‘pressing the switch’ and the bulb executes the command. As far as I know, Zigbee offers no way to diagnose why there is a difference. To the point of ‘flakey’ / buggy firmware, I agree there are probably devices on my network that have routing issues. Even the ‘better’ devices that do get firmware updates. I mean, you have to ask why a Philips Hue bulb that cost better than USD 50 when new, needs a dozen firmware upgrades over the time you have owned it…
One area that I do not see you mentioning, is the channel that your Zigbee network is on and whether that channel is overlayed by any Wifi channels? If true, this can be a big ‘random’ contributor to Zigbee networking issues. Again, above and beyond just making sure you Wifi and Zigbee frequencies are as far apart as possible, unless you have expensive spectrum analyzing tools, some rogue device that is randomly ‘splatting’ noise on top of your Zigbee frequency can be a problem. Even though Zigbee packets are small and infrequent, I have found that a Zigbee network can be easily disrupted, an operating microwave oven is a simple example. Good hunting!