Yes! 
Seeing LORA: recv_rpl: 0xd above, which is a newline, I wondered what would happen if I always first read any pending RX from the LoRa module’s UART before sending new commands.
Guess what: in two occasions there was indeed such excessive newline pending in the receive buffer while the firmware was about to send a new command. Even better: discarding those makes the LoRa configuration complete without any error. Next, the activation simply succeeds and it even downloads new firmware, which obviously no longer includes my fixes, but somehow is not booted (yet) as the SD card with my own firmware is still in the gateway, my own firmware is loaded again after the reboot. So, using my own firmware it’s now activated and happily forwarding packets…! And the debug logs no longer show garbage either.
To be continued, but: in my case the reboot loop during activation is apparently totally fixable by just using new firmware.
And I’m not even alarmed by this weird double occurrence of frame counter 9, on two frequencies…
(Well, I am…)