I tried a different approach and flashed FritzOS on the 7520 which should result in less confound results as the same hardware is used. Unfortunately, the FB ran out of RAM after about 8 h so the results were unrecoverably lost (can't save when there is an error in go-dsl). However, I observed nearly no errors both near and far at any point in time in go-dsl. Furthermore, FritzOS also did not report any line errors. This evidently leads to the conclusion that something in the way OpenWrt handles i.e. vectoring.
I found this (Vectoring on Lantiq VRX200 / VR9 - missing callback for sending error samples) post from @janh which points out that with the smaller modem (VRX200), there have been issues with deteriorating lines in the past. The specific issue has already been fixed in OpenWrt but there may be something else related at play in my case. Could this be a configuration error by any chance, i. e. using ptm instead of atm or using a wrong tone? It almost looks like there are communication errors due to incompatibilities.
The firmware files were all taken from the latest (v7.57) Fritz FW:
aca_fw.bin
ppe_fw.bin
vr11-B-dsl.bin
PS: I will run go-dsl again with FritzOS via -web. However, I want to refrain from using OpenWrt for a few days or weeks because I do not want my bandwidth to go down even further. It should recover over time due to the bandwidth management running on the DSLAM. I would really like to make this work (and it should, as @elder_tinkerer runs the same box without my issues), but if there is no chance of fixing the problem, I could try setting up the other FB 7512 in (half/full) bridge mode and use OpenWrt on the 7520.