It's happening because something changed (or 'went wrong', depending on how you look at it) when the device got migrated to ath79. The actual question is: did your original MAC match the vendor firmware's? Or is this the 'right' one? That's what you should find out. If you still got the box, with a bit of luck the MAC might still be on it. Often the MAC is extracted from the flash somewhere and then extrapolated (incremented, often) for the wireless interfaces.
eth0: DC:**:**:**:**:*D (same as eth0 on backfire)
eth1: DE:**:**:**:**:*D (not the same there was on backfire, it's prefixed with DE instead of DC, it ends in D instead of E)
eth0: DE:**:**:**:**:*D (corresponds to eth1 on ar71xx of 19.07)
eth1: DC:**:**:**:**:*D (corresponds to eth0 on ar71xx of 19.07)
So in essence, it seems eth0 and eth1 are swapped on ath79.
This mac address change is a bit problematic for network management systems like OpenWISP which recognize the device based on the mac address of a specific interface, if this interface changes, the device is not recognized, which causes issues when upgrading.
The issues are easy to resolve for me, but may not be so obvious to regular OpenWISP users.
I am not sure if it can be changed again since this was released in OpenWRT 19.07. What do you think?
I am testing also other ath79 images for other hardware models and have not found the same issue so far, I will report if I find any other similar issue on other models.
The Ethernet interfaces being swapped is something that happens frequently in the transition to ath79. I suppose there's a technical explanation for that but I'm not familiar with how that works exactly. The question is not how to replicate the Backfire behaviour, but to match the addresses to the stock firmware.
The MAC addresses are extracted from locations on the flash defined in the DTS; the Nanostation's DTS points (amongst others) to the XM DTSI which defines this:
As you can see both MACs are extracted from the ART partition. It looks like that's the wrong match for the Nanostation M2 then, unless that's how the stock firmware behaves. Only someone who can run the stock firmware on the M2 can provide conclusive evidence on that. I checked the Wikidevi clones but they haven't logged the MAC addresses either...