It’s considered best practice for multipathed peers - the loopback/dummy will remain visible over other paths even if the main interface goes down and loses its IP.
We use it to present a single peer IP that may be facing multiple interfaces and peers or direct interface addresses that might change in future. It’s commonly paired with OSPF or (rarely) static routes to the dummy /32 on downstream peers. We’ll usually also set the router-id to the IP of the loopback so that everything is consistent for diagnostics.
It’s not an appropriate configuration for all use cases - you often want to predict your physical pathing, don’t just dump a bunch of peers into a mixed bag of L2 links and hope OSPF figures it all out - but it is a good best practice, very convenient for intercap interior routers, RR clusters and BGP sessions with multiple L2 paths between peers.
It will also regularly simplify peer-group configs and allow downstream peer configs to be copy-pasted, even across routers facing different direct interfaces.