As I do not plan to run any other VMs except for backup VM, i will kindly ask you to elaborate on "add coupling between the cluster (one of the backup sources) and the backup server (the target), which may become an issue on outages".
What kind of issue can arise on what kind of outages from this coupling?
You double the operating systems used, doubling outage risks from operating system issues (not a 100% true calculation, but to give you an idea).
Putting your Backup system, which may be the most important during an outage or other issues, inside a VM means that if the PVE host does not boot anymore you need to first debug that before getting even to start checking out the backup VM. Simpler setups, meaning less complexity below the backup service here, should be prioritized to minimize the things which can fail between you and the backup.
How does that differ from the same issue arising with usual PM hosts and their guests?
Other VMs on PVE hosts do not host backups. A similar coupling could be described with using a VM as a firewall/router for the whole system, also the physical PVE hosts it runs on. While it can work out, you have a bigger risk as on an outage you need to bootstrap the network from the outside, which maybe hard - especially if you could only connect over that (now dead) network.
Using it as a VM on a dedicated node, allows me to move the backup service to another (possibly bigger) node easily and to get another vote for my cluster quorum, where cluster is two production nodes (with replication) and third node for backup.
I'd then rather suggest installing Proxmox Backup Server and Proxmox VE alongside on the same host, without CT or VM separation. Here, if the host boot really fails, you could always just use a PBS ISO + a USB drive and install it there to get a running backup server with all your data (provided that was on another storage than the OS) again up and running.
If you need to move to a bigger server, you could then use the native remote sync feature of the PBS.
So, my point is, you won't gain actually something from using a VM for PBS in the described scenarios and there are some situations where it may even hurt.
If you'd use a redundant shared storage like a ceph setup just for the backup data, one could argue that having a VM, which then can be migrated efficiently and fast (no storage migration).
But, then you naturally would need to handle the extra complexity of a separate redundant storage setup, which can also bring complexity and bootstrap on outage issues with it.
TL;DR: it normally will work OK, but not provide many benefits. On the other hand, reducing layers for such critical things can help in some situations.
