Post History

It seems that what the linked article says is usually correct, about the interfaces which are part of the bridge not being supposed to have IPs themselves.
At least in the wanted subnet, I guess.

This seems a bit special here. I can't get the USB gadget stuff to work, i.e. any traffic between gadget and host, unless I assign the usb0 interfaces on both, gadget and host, an address in the 169.254.*.* net (or rather, let it auto assign that).
So what I now do is, let the ethernet gadget stuff self-assign such addresses as it may, but assign no further (actually targeted subnet) address to usb0 nor eth0 on the host - only the bridge that connectes those two has such an address. On the gadget, the usb0 does get an additional IP which lies in the targeted ("intranet") subnet, though - as that's what's being bridged to the outside, on the host.
That works.

I have found no reference that states: that, and why, interfaces involved in a USB ethernet gadget connection always need an IP (let alone in a special subnet).
So this, unfortunately, is not quite a canonical answer.
All I can say is: "this works for me".